xref: /openbmc/linux/fs/namespace.c (revision 8ee90c5c)
1 /*
2  *  linux/fs/namespace.c
3  *
4  * (C) Copyright Al Viro 2000, 2001
5  *	Released under GPL v2.
6  *
7  * Based on code from fs/super.c, copyright Linus Torvalds and others.
8  * Heavily rewritten.
9  */
10 
11 #include <linux/syscalls.h>
12 #include <linux/export.h>
13 #include <linux/capability.h>
14 #include <linux/mnt_namespace.h>
15 #include <linux/user_namespace.h>
16 #include <linux/namei.h>
17 #include <linux/security.h>
18 #include <linux/cred.h>
19 #include <linux/idr.h>
20 #include <linux/init.h>		/* init_rootfs */
21 #include <linux/fs_struct.h>	/* get_fs_root et.al. */
22 #include <linux/fsnotify.h>	/* fsnotify_vfsmount_delete */
23 #include <linux/uaccess.h>
24 #include <linux/proc_ns.h>
25 #include <linux/magic.h>
26 #include <linux/bootmem.h>
27 #include <linux/task_work.h>
28 #include <linux/sched/task.h>
29 
30 #include "pnode.h"
31 #include "internal.h"
32 
33 /* Maximum number of mounts in a mount namespace */
34 unsigned int sysctl_mount_max __read_mostly = 100000;
35 
36 static unsigned int m_hash_mask __read_mostly;
37 static unsigned int m_hash_shift __read_mostly;
38 static unsigned int mp_hash_mask __read_mostly;
39 static unsigned int mp_hash_shift __read_mostly;
40 
41 static __initdata unsigned long mhash_entries;
42 static int __init set_mhash_entries(char *str)
43 {
44 	if (!str)
45 		return 0;
46 	mhash_entries = simple_strtoul(str, &str, 0);
47 	return 1;
48 }
49 __setup("mhash_entries=", set_mhash_entries);
50 
51 static __initdata unsigned long mphash_entries;
52 static int __init set_mphash_entries(char *str)
53 {
54 	if (!str)
55 		return 0;
56 	mphash_entries = simple_strtoul(str, &str, 0);
57 	return 1;
58 }
59 __setup("mphash_entries=", set_mphash_entries);
60 
61 static u64 event;
62 static DEFINE_IDA(mnt_id_ida);
63 static DEFINE_IDA(mnt_group_ida);
64 static DEFINE_SPINLOCK(mnt_id_lock);
65 static int mnt_id_start = 0;
66 static int mnt_group_start = 1;
67 
68 static struct hlist_head *mount_hashtable __read_mostly;
69 static struct hlist_head *mountpoint_hashtable __read_mostly;
70 static struct kmem_cache *mnt_cache __read_mostly;
71 static DECLARE_RWSEM(namespace_sem);
72 
73 /* /sys/fs */
74 struct kobject *fs_kobj;
75 EXPORT_SYMBOL_GPL(fs_kobj);
76 
77 /*
78  * vfsmount lock may be taken for read to prevent changes to the
79  * vfsmount hash, ie. during mountpoint lookups or walking back
80  * up the tree.
81  *
82  * It should be taken for write in all cases where the vfsmount
83  * tree or hash is modified or when a vfsmount structure is modified.
84  */
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
86 
87 static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
88 {
89 	unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
90 	tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
91 	tmp = tmp + (tmp >> m_hash_shift);
92 	return &mount_hashtable[tmp & m_hash_mask];
93 }
94 
95 static inline struct hlist_head *mp_hash(struct dentry *dentry)
96 {
97 	unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
98 	tmp = tmp + (tmp >> mp_hash_shift);
99 	return &mountpoint_hashtable[tmp & mp_hash_mask];
100 }
101 
102 static int mnt_alloc_id(struct mount *mnt)
103 {
104 	int res;
105 
106 retry:
107 	ida_pre_get(&mnt_id_ida, GFP_KERNEL);
108 	spin_lock(&mnt_id_lock);
109 	res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
110 	if (!res)
111 		mnt_id_start = mnt->mnt_id + 1;
112 	spin_unlock(&mnt_id_lock);
113 	if (res == -EAGAIN)
114 		goto retry;
115 
116 	return res;
117 }
118 
119 static void mnt_free_id(struct mount *mnt)
120 {
121 	int id = mnt->mnt_id;
122 	spin_lock(&mnt_id_lock);
123 	ida_remove(&mnt_id_ida, id);
124 	if (mnt_id_start > id)
125 		mnt_id_start = id;
126 	spin_unlock(&mnt_id_lock);
127 }
128 
129 /*
130  * Allocate a new peer group ID
131  *
132  * mnt_group_ida is protected by namespace_sem
133  */
134 static int mnt_alloc_group_id(struct mount *mnt)
135 {
136 	int res;
137 
138 	if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
139 		return -ENOMEM;
140 
141 	res = ida_get_new_above(&mnt_group_ida,
142 				mnt_group_start,
143 				&mnt->mnt_group_id);
144 	if (!res)
145 		mnt_group_start = mnt->mnt_group_id + 1;
146 
147 	return res;
148 }
149 
150 /*
151  * Release a peer group ID
152  */
153 void mnt_release_group_id(struct mount *mnt)
154 {
155 	int id = mnt->mnt_group_id;
156 	ida_remove(&mnt_group_ida, id);
157 	if (mnt_group_start > id)
158 		mnt_group_start = id;
159 	mnt->mnt_group_id = 0;
160 }
161 
162 /*
163  * vfsmount lock must be held for read
164  */
165 static inline void mnt_add_count(struct mount *mnt, int n)
166 {
167 #ifdef CONFIG_SMP
168 	this_cpu_add(mnt->mnt_pcp->mnt_count, n);
169 #else
170 	preempt_disable();
171 	mnt->mnt_count += n;
172 	preempt_enable();
173 #endif
174 }
175 
176 /*
177  * vfsmount lock must be held for write
178  */
179 unsigned int mnt_get_count(struct mount *mnt)
180 {
181 #ifdef CONFIG_SMP
182 	unsigned int count = 0;
183 	int cpu;
184 
185 	for_each_possible_cpu(cpu) {
186 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
187 	}
188 
189 	return count;
190 #else
191 	return mnt->mnt_count;
192 #endif
193 }
194 
195 static void drop_mountpoint(struct fs_pin *p)
196 {
197 	struct mount *m = container_of(p, struct mount, mnt_umount);
198 	dput(m->mnt_ex_mountpoint);
199 	pin_remove(p);
200 	mntput(&m->mnt);
201 }
202 
203 static struct mount *alloc_vfsmnt(const char *name)
204 {
205 	struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
206 	if (mnt) {
207 		int err;
208 
209 		err = mnt_alloc_id(mnt);
210 		if (err)
211 			goto out_free_cache;
212 
213 		if (name) {
214 			mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
215 			if (!mnt->mnt_devname)
216 				goto out_free_id;
217 		}
218 
219 #ifdef CONFIG_SMP
220 		mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
221 		if (!mnt->mnt_pcp)
222 			goto out_free_devname;
223 
224 		this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
225 #else
226 		mnt->mnt_count = 1;
227 		mnt->mnt_writers = 0;
228 #endif
229 
230 		INIT_HLIST_NODE(&mnt->mnt_hash);
231 		INIT_LIST_HEAD(&mnt->mnt_child);
232 		INIT_LIST_HEAD(&mnt->mnt_mounts);
233 		INIT_LIST_HEAD(&mnt->mnt_list);
234 		INIT_LIST_HEAD(&mnt->mnt_expire);
235 		INIT_LIST_HEAD(&mnt->mnt_share);
236 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
237 		INIT_LIST_HEAD(&mnt->mnt_slave);
238 		INIT_HLIST_NODE(&mnt->mnt_mp_list);
239 		INIT_LIST_HEAD(&mnt->mnt_umounting);
240 		init_fs_pin(&mnt->mnt_umount, drop_mountpoint);
241 	}
242 	return mnt;
243 
244 #ifdef CONFIG_SMP
245 out_free_devname:
246 	kfree_const(mnt->mnt_devname);
247 #endif
248 out_free_id:
249 	mnt_free_id(mnt);
250 out_free_cache:
251 	kmem_cache_free(mnt_cache, mnt);
252 	return NULL;
253 }
254 
255 /*
256  * Most r/o checks on a fs are for operations that take
257  * discrete amounts of time, like a write() or unlink().
258  * We must keep track of when those operations start
259  * (for permission checks) and when they end, so that
260  * we can determine when writes are able to occur to
261  * a filesystem.
262  */
263 /*
264  * __mnt_is_readonly: check whether a mount is read-only
265  * @mnt: the mount to check for its write status
266  *
267  * This shouldn't be used directly ouside of the VFS.
268  * It does not guarantee that the filesystem will stay
269  * r/w, just that it is right *now*.  This can not and
270  * should not be used in place of IS_RDONLY(inode).
271  * mnt_want/drop_write() will _keep_ the filesystem
272  * r/w.
273  */
274 int __mnt_is_readonly(struct vfsmount *mnt)
275 {
276 	if (mnt->mnt_flags & MNT_READONLY)
277 		return 1;
278 	if (sb_rdonly(mnt->mnt_sb))
279 		return 1;
280 	return 0;
281 }
282 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
283 
284 static inline void mnt_inc_writers(struct mount *mnt)
285 {
286 #ifdef CONFIG_SMP
287 	this_cpu_inc(mnt->mnt_pcp->mnt_writers);
288 #else
289 	mnt->mnt_writers++;
290 #endif
291 }
292 
293 static inline void mnt_dec_writers(struct mount *mnt)
294 {
295 #ifdef CONFIG_SMP
296 	this_cpu_dec(mnt->mnt_pcp->mnt_writers);
297 #else
298 	mnt->mnt_writers--;
299 #endif
300 }
301 
302 static unsigned int mnt_get_writers(struct mount *mnt)
303 {
304 #ifdef CONFIG_SMP
305 	unsigned int count = 0;
306 	int cpu;
307 
308 	for_each_possible_cpu(cpu) {
309 		count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
310 	}
311 
312 	return count;
313 #else
314 	return mnt->mnt_writers;
315 #endif
316 }
317 
318 static int mnt_is_readonly(struct vfsmount *mnt)
319 {
320 	if (mnt->mnt_sb->s_readonly_remount)
321 		return 1;
322 	/* Order wrt setting s_flags/s_readonly_remount in do_remount() */
323 	smp_rmb();
324 	return __mnt_is_readonly(mnt);
325 }
326 
327 /*
328  * Most r/o & frozen checks on a fs are for operations that take discrete
329  * amounts of time, like a write() or unlink().  We must keep track of when
330  * those operations start (for permission checks) and when they end, so that we
331  * can determine when writes are able to occur to a filesystem.
332  */
333 /**
334  * __mnt_want_write - get write access to a mount without freeze protection
335  * @m: the mount on which to take a write
336  *
337  * This tells the low-level filesystem that a write is about to be performed to
338  * it, and makes sure that writes are allowed (mnt it read-write) before
339  * returning success. This operation does not protect against filesystem being
340  * frozen. When the write operation is finished, __mnt_drop_write() must be
341  * called. This is effectively a refcount.
342  */
343 int __mnt_want_write(struct vfsmount *m)
344 {
345 	struct mount *mnt = real_mount(m);
346 	int ret = 0;
347 
348 	preempt_disable();
349 	mnt_inc_writers(mnt);
350 	/*
351 	 * The store to mnt_inc_writers must be visible before we pass
352 	 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
353 	 * incremented count after it has set MNT_WRITE_HOLD.
354 	 */
355 	smp_mb();
356 	while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
357 		cpu_relax();
358 	/*
359 	 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
360 	 * be set to match its requirements. So we must not load that until
361 	 * MNT_WRITE_HOLD is cleared.
362 	 */
363 	smp_rmb();
364 	if (mnt_is_readonly(m)) {
365 		mnt_dec_writers(mnt);
366 		ret = -EROFS;
367 	}
368 	preempt_enable();
369 
370 	return ret;
371 }
372 
373 /**
374  * mnt_want_write - get write access to a mount
375  * @m: the mount on which to take a write
376  *
377  * This tells the low-level filesystem that a write is about to be performed to
378  * it, and makes sure that writes are allowed (mount is read-write, filesystem
379  * is not frozen) before returning success.  When the write operation is
380  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
381  */
382 int mnt_want_write(struct vfsmount *m)
383 {
384 	int ret;
385 
386 	sb_start_write(m->mnt_sb);
387 	ret = __mnt_want_write(m);
388 	if (ret)
389 		sb_end_write(m->mnt_sb);
390 	return ret;
391 }
392 EXPORT_SYMBOL_GPL(mnt_want_write);
393 
394 /**
395  * mnt_clone_write - get write access to a mount
396  * @mnt: the mount on which to take a write
397  *
398  * This is effectively like mnt_want_write, except
399  * it must only be used to take an extra write reference
400  * on a mountpoint that we already know has a write reference
401  * on it. This allows some optimisation.
402  *
403  * After finished, mnt_drop_write must be called as usual to
404  * drop the reference.
405  */
406 int mnt_clone_write(struct vfsmount *mnt)
407 {
408 	/* superblock may be r/o */
409 	if (__mnt_is_readonly(mnt))
410 		return -EROFS;
411 	preempt_disable();
412 	mnt_inc_writers(real_mount(mnt));
413 	preempt_enable();
414 	return 0;
415 }
416 EXPORT_SYMBOL_GPL(mnt_clone_write);
417 
418 /**
419  * __mnt_want_write_file - get write access to a file's mount
420  * @file: the file who's mount on which to take a write
421  *
422  * This is like __mnt_want_write, but it takes a file and can
423  * do some optimisations if the file is open for write already
424  */
425 int __mnt_want_write_file(struct file *file)
426 {
427 	if (!(file->f_mode & FMODE_WRITER))
428 		return __mnt_want_write(file->f_path.mnt);
429 	else
430 		return mnt_clone_write(file->f_path.mnt);
431 }
432 
433 /**
434  * mnt_want_write_file_path - get write access to a file's mount
435  * @file: the file who's mount on which to take a write
436  *
437  * This is like mnt_want_write, but it takes a file and can
438  * do some optimisations if the file is open for write already
439  *
440  * Called by the vfs for cases when we have an open file at hand, but will do an
441  * inode operation on it (important distinction for files opened on overlayfs,
442  * since the file operations will come from the real underlying file, while
443  * inode operations come from the overlay).
444  */
445 int mnt_want_write_file_path(struct file *file)
446 {
447 	int ret;
448 
449 	sb_start_write(file->f_path.mnt->mnt_sb);
450 	ret = __mnt_want_write_file(file);
451 	if (ret)
452 		sb_end_write(file->f_path.mnt->mnt_sb);
453 	return ret;
454 }
455 
456 static inline int may_write_real(struct file *file)
457 {
458 	struct dentry *dentry = file->f_path.dentry;
459 	struct dentry *upperdentry;
460 
461 	/* Writable file? */
462 	if (file->f_mode & FMODE_WRITER)
463 		return 0;
464 
465 	/* Not overlayfs? */
466 	if (likely(!(dentry->d_flags & DCACHE_OP_REAL)))
467 		return 0;
468 
469 	/* File refers to upper, writable layer? */
470 	upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER);
471 	if (upperdentry &&
472 	    (file_inode(file) == d_inode(upperdentry) ||
473 	     file_inode(file) == d_inode(dentry)))
474 		return 0;
475 
476 	/* Lower layer: can't write to real file, sorry... */
477 	return -EPERM;
478 }
479 
480 /**
481  * mnt_want_write_file - get write access to a file's mount
482  * @file: the file who's mount on which to take a write
483  *
484  * This is like mnt_want_write, but it takes a file and can
485  * do some optimisations if the file is open for write already
486  *
487  * Mostly called by filesystems from their ioctl operation before performing
488  * modification.  On overlayfs this needs to check if the file is on a read-only
489  * lower layer and deny access in that case.
490  */
491 int mnt_want_write_file(struct file *file)
492 {
493 	int ret;
494 
495 	ret = may_write_real(file);
496 	if (!ret) {
497 		sb_start_write(file_inode(file)->i_sb);
498 		ret = __mnt_want_write_file(file);
499 		if (ret)
500 			sb_end_write(file_inode(file)->i_sb);
501 	}
502 	return ret;
503 }
504 EXPORT_SYMBOL_GPL(mnt_want_write_file);
505 
506 /**
507  * __mnt_drop_write - give up write access to a mount
508  * @mnt: the mount on which to give up write access
509  *
510  * Tells the low-level filesystem that we are done
511  * performing writes to it.  Must be matched with
512  * __mnt_want_write() call above.
513  */
514 void __mnt_drop_write(struct vfsmount *mnt)
515 {
516 	preempt_disable();
517 	mnt_dec_writers(real_mount(mnt));
518 	preempt_enable();
519 }
520 
521 /**
522  * mnt_drop_write - give up write access to a mount
523  * @mnt: the mount on which to give up write access
524  *
525  * Tells the low-level filesystem that we are done performing writes to it and
526  * also allows filesystem to be frozen again.  Must be matched with
527  * mnt_want_write() call above.
528  */
529 void mnt_drop_write(struct vfsmount *mnt)
530 {
531 	__mnt_drop_write(mnt);
532 	sb_end_write(mnt->mnt_sb);
533 }
534 EXPORT_SYMBOL_GPL(mnt_drop_write);
535 
536 void __mnt_drop_write_file(struct file *file)
537 {
538 	__mnt_drop_write(file->f_path.mnt);
539 }
540 
541 void mnt_drop_write_file_path(struct file *file)
542 {
543 	mnt_drop_write(file->f_path.mnt);
544 }
545 
546 void mnt_drop_write_file(struct file *file)
547 {
548 	__mnt_drop_write(file->f_path.mnt);
549 	sb_end_write(file_inode(file)->i_sb);
550 }
551 EXPORT_SYMBOL(mnt_drop_write_file);
552 
553 static int mnt_make_readonly(struct mount *mnt)
554 {
555 	int ret = 0;
556 
557 	lock_mount_hash();
558 	mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
559 	/*
560 	 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
561 	 * should be visible before we do.
562 	 */
563 	smp_mb();
564 
565 	/*
566 	 * With writers on hold, if this value is zero, then there are
567 	 * definitely no active writers (although held writers may subsequently
568 	 * increment the count, they'll have to wait, and decrement it after
569 	 * seeing MNT_READONLY).
570 	 *
571 	 * It is OK to have counter incremented on one CPU and decremented on
572 	 * another: the sum will add up correctly. The danger would be when we
573 	 * sum up each counter, if we read a counter before it is incremented,
574 	 * but then read another CPU's count which it has been subsequently
575 	 * decremented from -- we would see more decrements than we should.
576 	 * MNT_WRITE_HOLD protects against this scenario, because
577 	 * mnt_want_write first increments count, then smp_mb, then spins on
578 	 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
579 	 * we're counting up here.
580 	 */
581 	if (mnt_get_writers(mnt) > 0)
582 		ret = -EBUSY;
583 	else
584 		mnt->mnt.mnt_flags |= MNT_READONLY;
585 	/*
586 	 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
587 	 * that become unheld will see MNT_READONLY.
588 	 */
589 	smp_wmb();
590 	mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
591 	unlock_mount_hash();
592 	return ret;
593 }
594 
595 static void __mnt_unmake_readonly(struct mount *mnt)
596 {
597 	lock_mount_hash();
598 	mnt->mnt.mnt_flags &= ~MNT_READONLY;
599 	unlock_mount_hash();
600 }
601 
602 int sb_prepare_remount_readonly(struct super_block *sb)
603 {
604 	struct mount *mnt;
605 	int err = 0;
606 
607 	/* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
608 	if (atomic_long_read(&sb->s_remove_count))
609 		return -EBUSY;
610 
611 	lock_mount_hash();
612 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
613 		if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
614 			mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
615 			smp_mb();
616 			if (mnt_get_writers(mnt) > 0) {
617 				err = -EBUSY;
618 				break;
619 			}
620 		}
621 	}
622 	if (!err && atomic_long_read(&sb->s_remove_count))
623 		err = -EBUSY;
624 
625 	if (!err) {
626 		sb->s_readonly_remount = 1;
627 		smp_wmb();
628 	}
629 	list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
630 		if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
631 			mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
632 	}
633 	unlock_mount_hash();
634 
635 	return err;
636 }
637 
638 static void free_vfsmnt(struct mount *mnt)
639 {
640 	kfree_const(mnt->mnt_devname);
641 #ifdef CONFIG_SMP
642 	free_percpu(mnt->mnt_pcp);
643 #endif
644 	kmem_cache_free(mnt_cache, mnt);
645 }
646 
647 static void delayed_free_vfsmnt(struct rcu_head *head)
648 {
649 	free_vfsmnt(container_of(head, struct mount, mnt_rcu));
650 }
651 
652 /* call under rcu_read_lock */
653 int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
654 {
655 	struct mount *mnt;
656 	if (read_seqretry(&mount_lock, seq))
657 		return 1;
658 	if (bastard == NULL)
659 		return 0;
660 	mnt = real_mount(bastard);
661 	mnt_add_count(mnt, 1);
662 	if (likely(!read_seqretry(&mount_lock, seq)))
663 		return 0;
664 	if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
665 		mnt_add_count(mnt, -1);
666 		return 1;
667 	}
668 	return -1;
669 }
670 
671 /* call under rcu_read_lock */
672 bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
673 {
674 	int res = __legitimize_mnt(bastard, seq);
675 	if (likely(!res))
676 		return true;
677 	if (unlikely(res < 0)) {
678 		rcu_read_unlock();
679 		mntput(bastard);
680 		rcu_read_lock();
681 	}
682 	return false;
683 }
684 
685 /*
686  * find the first mount at @dentry on vfsmount @mnt.
687  * call under rcu_read_lock()
688  */
689 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
690 {
691 	struct hlist_head *head = m_hash(mnt, dentry);
692 	struct mount *p;
693 
694 	hlist_for_each_entry_rcu(p, head, mnt_hash)
695 		if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
696 			return p;
697 	return NULL;
698 }
699 
700 /*
701  * lookup_mnt - Return the first child mount mounted at path
702  *
703  * "First" means first mounted chronologically.  If you create the
704  * following mounts:
705  *
706  * mount /dev/sda1 /mnt
707  * mount /dev/sda2 /mnt
708  * mount /dev/sda3 /mnt
709  *
710  * Then lookup_mnt() on the base /mnt dentry in the root mount will
711  * return successively the root dentry and vfsmount of /dev/sda1, then
712  * /dev/sda2, then /dev/sda3, then NULL.
713  *
714  * lookup_mnt takes a reference to the found vfsmount.
715  */
716 struct vfsmount *lookup_mnt(const struct path *path)
717 {
718 	struct mount *child_mnt;
719 	struct vfsmount *m;
720 	unsigned seq;
721 
722 	rcu_read_lock();
723 	do {
724 		seq = read_seqbegin(&mount_lock);
725 		child_mnt = __lookup_mnt(path->mnt, path->dentry);
726 		m = child_mnt ? &child_mnt->mnt : NULL;
727 	} while (!legitimize_mnt(m, seq));
728 	rcu_read_unlock();
729 	return m;
730 }
731 
732 /*
733  * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
734  *                         current mount namespace.
735  *
736  * The common case is dentries are not mountpoints at all and that
737  * test is handled inline.  For the slow case when we are actually
738  * dealing with a mountpoint of some kind, walk through all of the
739  * mounts in the current mount namespace and test to see if the dentry
740  * is a mountpoint.
741  *
742  * The mount_hashtable is not usable in the context because we
743  * need to identify all mounts that may be in the current mount
744  * namespace not just a mount that happens to have some specified
745  * parent mount.
746  */
747 bool __is_local_mountpoint(struct dentry *dentry)
748 {
749 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
750 	struct mount *mnt;
751 	bool is_covered = false;
752 
753 	if (!d_mountpoint(dentry))
754 		goto out;
755 
756 	down_read(&namespace_sem);
757 	list_for_each_entry(mnt, &ns->list, mnt_list) {
758 		is_covered = (mnt->mnt_mountpoint == dentry);
759 		if (is_covered)
760 			break;
761 	}
762 	up_read(&namespace_sem);
763 out:
764 	return is_covered;
765 }
766 
767 static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
768 {
769 	struct hlist_head *chain = mp_hash(dentry);
770 	struct mountpoint *mp;
771 
772 	hlist_for_each_entry(mp, chain, m_hash) {
773 		if (mp->m_dentry == dentry) {
774 			/* might be worth a WARN_ON() */
775 			if (d_unlinked(dentry))
776 				return ERR_PTR(-ENOENT);
777 			mp->m_count++;
778 			return mp;
779 		}
780 	}
781 	return NULL;
782 }
783 
784 static struct mountpoint *get_mountpoint(struct dentry *dentry)
785 {
786 	struct mountpoint *mp, *new = NULL;
787 	int ret;
788 
789 	if (d_mountpoint(dentry)) {
790 mountpoint:
791 		read_seqlock_excl(&mount_lock);
792 		mp = lookup_mountpoint(dentry);
793 		read_sequnlock_excl(&mount_lock);
794 		if (mp)
795 			goto done;
796 	}
797 
798 	if (!new)
799 		new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
800 	if (!new)
801 		return ERR_PTR(-ENOMEM);
802 
803 
804 	/* Exactly one processes may set d_mounted */
805 	ret = d_set_mounted(dentry);
806 
807 	/* Someone else set d_mounted? */
808 	if (ret == -EBUSY)
809 		goto mountpoint;
810 
811 	/* The dentry is not available as a mountpoint? */
812 	mp = ERR_PTR(ret);
813 	if (ret)
814 		goto done;
815 
816 	/* Add the new mountpoint to the hash table */
817 	read_seqlock_excl(&mount_lock);
818 	new->m_dentry = dentry;
819 	new->m_count = 1;
820 	hlist_add_head(&new->m_hash, mp_hash(dentry));
821 	INIT_HLIST_HEAD(&new->m_list);
822 	read_sequnlock_excl(&mount_lock);
823 
824 	mp = new;
825 	new = NULL;
826 done:
827 	kfree(new);
828 	return mp;
829 }
830 
831 static void put_mountpoint(struct mountpoint *mp)
832 {
833 	if (!--mp->m_count) {
834 		struct dentry *dentry = mp->m_dentry;
835 		BUG_ON(!hlist_empty(&mp->m_list));
836 		spin_lock(&dentry->d_lock);
837 		dentry->d_flags &= ~DCACHE_MOUNTED;
838 		spin_unlock(&dentry->d_lock);
839 		hlist_del(&mp->m_hash);
840 		kfree(mp);
841 	}
842 }
843 
844 static inline int check_mnt(struct mount *mnt)
845 {
846 	return mnt->mnt_ns == current->nsproxy->mnt_ns;
847 }
848 
849 /*
850  * vfsmount lock must be held for write
851  */
852 static void touch_mnt_namespace(struct mnt_namespace *ns)
853 {
854 	if (ns) {
855 		ns->event = ++event;
856 		wake_up_interruptible(&ns->poll);
857 	}
858 }
859 
860 /*
861  * vfsmount lock must be held for write
862  */
863 static void __touch_mnt_namespace(struct mnt_namespace *ns)
864 {
865 	if (ns && ns->event != event) {
866 		ns->event = event;
867 		wake_up_interruptible(&ns->poll);
868 	}
869 }
870 
871 /*
872  * vfsmount lock must be held for write
873  */
874 static void unhash_mnt(struct mount *mnt)
875 {
876 	mnt->mnt_parent = mnt;
877 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
878 	list_del_init(&mnt->mnt_child);
879 	hlist_del_init_rcu(&mnt->mnt_hash);
880 	hlist_del_init(&mnt->mnt_mp_list);
881 	put_mountpoint(mnt->mnt_mp);
882 	mnt->mnt_mp = NULL;
883 }
884 
885 /*
886  * vfsmount lock must be held for write
887  */
888 static void detach_mnt(struct mount *mnt, struct path *old_path)
889 {
890 	old_path->dentry = mnt->mnt_mountpoint;
891 	old_path->mnt = &mnt->mnt_parent->mnt;
892 	unhash_mnt(mnt);
893 }
894 
895 /*
896  * vfsmount lock must be held for write
897  */
898 static void umount_mnt(struct mount *mnt)
899 {
900 	/* old mountpoint will be dropped when we can do that */
901 	mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint;
902 	unhash_mnt(mnt);
903 }
904 
905 /*
906  * vfsmount lock must be held for write
907  */
908 void mnt_set_mountpoint(struct mount *mnt,
909 			struct mountpoint *mp,
910 			struct mount *child_mnt)
911 {
912 	mp->m_count++;
913 	mnt_add_count(mnt, 1);	/* essentially, that's mntget */
914 	child_mnt->mnt_mountpoint = dget(mp->m_dentry);
915 	child_mnt->mnt_parent = mnt;
916 	child_mnt->mnt_mp = mp;
917 	hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
918 }
919 
920 static void __attach_mnt(struct mount *mnt, struct mount *parent)
921 {
922 	hlist_add_head_rcu(&mnt->mnt_hash,
923 			   m_hash(&parent->mnt, mnt->mnt_mountpoint));
924 	list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
925 }
926 
927 /*
928  * vfsmount lock must be held for write
929  */
930 static void attach_mnt(struct mount *mnt,
931 			struct mount *parent,
932 			struct mountpoint *mp)
933 {
934 	mnt_set_mountpoint(parent, mp, mnt);
935 	__attach_mnt(mnt, parent);
936 }
937 
938 void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
939 {
940 	struct mountpoint *old_mp = mnt->mnt_mp;
941 	struct dentry *old_mountpoint = mnt->mnt_mountpoint;
942 	struct mount *old_parent = mnt->mnt_parent;
943 
944 	list_del_init(&mnt->mnt_child);
945 	hlist_del_init(&mnt->mnt_mp_list);
946 	hlist_del_init_rcu(&mnt->mnt_hash);
947 
948 	attach_mnt(mnt, parent, mp);
949 
950 	put_mountpoint(old_mp);
951 
952 	/*
953 	 * Safely avoid even the suggestion this code might sleep or
954 	 * lock the mount hash by taking advantage of the knowledge that
955 	 * mnt_change_mountpoint will not release the final reference
956 	 * to a mountpoint.
957 	 *
958 	 * During mounting, the mount passed in as the parent mount will
959 	 * continue to use the old mountpoint and during unmounting, the
960 	 * old mountpoint will continue to exist until namespace_unlock,
961 	 * which happens well after mnt_change_mountpoint.
962 	 */
963 	spin_lock(&old_mountpoint->d_lock);
964 	old_mountpoint->d_lockref.count--;
965 	spin_unlock(&old_mountpoint->d_lock);
966 
967 	mnt_add_count(old_parent, -1);
968 }
969 
970 /*
971  * vfsmount lock must be held for write
972  */
973 static void commit_tree(struct mount *mnt)
974 {
975 	struct mount *parent = mnt->mnt_parent;
976 	struct mount *m;
977 	LIST_HEAD(head);
978 	struct mnt_namespace *n = parent->mnt_ns;
979 
980 	BUG_ON(parent == mnt);
981 
982 	list_add_tail(&head, &mnt->mnt_list);
983 	list_for_each_entry(m, &head, mnt_list)
984 		m->mnt_ns = n;
985 
986 	list_splice(&head, n->list.prev);
987 
988 	n->mounts += n->pending_mounts;
989 	n->pending_mounts = 0;
990 
991 	__attach_mnt(mnt, parent);
992 	touch_mnt_namespace(n);
993 }
994 
995 static struct mount *next_mnt(struct mount *p, struct mount *root)
996 {
997 	struct list_head *next = p->mnt_mounts.next;
998 	if (next == &p->mnt_mounts) {
999 		while (1) {
1000 			if (p == root)
1001 				return NULL;
1002 			next = p->mnt_child.next;
1003 			if (next != &p->mnt_parent->mnt_mounts)
1004 				break;
1005 			p = p->mnt_parent;
1006 		}
1007 	}
1008 	return list_entry(next, struct mount, mnt_child);
1009 }
1010 
1011 static struct mount *skip_mnt_tree(struct mount *p)
1012 {
1013 	struct list_head *prev = p->mnt_mounts.prev;
1014 	while (prev != &p->mnt_mounts) {
1015 		p = list_entry(prev, struct mount, mnt_child);
1016 		prev = p->mnt_mounts.prev;
1017 	}
1018 	return p;
1019 }
1020 
1021 struct vfsmount *
1022 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
1023 {
1024 	struct mount *mnt;
1025 	struct dentry *root;
1026 
1027 	if (!type)
1028 		return ERR_PTR(-ENODEV);
1029 
1030 	mnt = alloc_vfsmnt(name);
1031 	if (!mnt)
1032 		return ERR_PTR(-ENOMEM);
1033 
1034 	if (flags & SB_KERNMOUNT)
1035 		mnt->mnt.mnt_flags = MNT_INTERNAL;
1036 
1037 	root = mount_fs(type, flags, name, data);
1038 	if (IS_ERR(root)) {
1039 		mnt_free_id(mnt);
1040 		free_vfsmnt(mnt);
1041 		return ERR_CAST(root);
1042 	}
1043 
1044 	mnt->mnt.mnt_root = root;
1045 	mnt->mnt.mnt_sb = root->d_sb;
1046 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1047 	mnt->mnt_parent = mnt;
1048 	lock_mount_hash();
1049 	list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
1050 	unlock_mount_hash();
1051 	return &mnt->mnt;
1052 }
1053 EXPORT_SYMBOL_GPL(vfs_kern_mount);
1054 
1055 struct vfsmount *
1056 vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1057 	     const char *name, void *data)
1058 {
1059 	/* Until it is worked out how to pass the user namespace
1060 	 * through from the parent mount to the submount don't support
1061 	 * unprivileged mounts with submounts.
1062 	 */
1063 	if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1064 		return ERR_PTR(-EPERM);
1065 
1066 	return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1067 }
1068 EXPORT_SYMBOL_GPL(vfs_submount);
1069 
1070 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1071 					int flag)
1072 {
1073 	struct super_block *sb = old->mnt.mnt_sb;
1074 	struct mount *mnt;
1075 	int err;
1076 
1077 	mnt = alloc_vfsmnt(old->mnt_devname);
1078 	if (!mnt)
1079 		return ERR_PTR(-ENOMEM);
1080 
1081 	if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1082 		mnt->mnt_group_id = 0; /* not a peer of original */
1083 	else
1084 		mnt->mnt_group_id = old->mnt_group_id;
1085 
1086 	if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1087 		err = mnt_alloc_group_id(mnt);
1088 		if (err)
1089 			goto out_free;
1090 	}
1091 
1092 	mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED);
1093 	/* Don't allow unprivileged users to change mount flags */
1094 	if (flag & CL_UNPRIVILEGED) {
1095 		mnt->mnt.mnt_flags |= MNT_LOCK_ATIME;
1096 
1097 		if (mnt->mnt.mnt_flags & MNT_READONLY)
1098 			mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
1099 
1100 		if (mnt->mnt.mnt_flags & MNT_NODEV)
1101 			mnt->mnt.mnt_flags |= MNT_LOCK_NODEV;
1102 
1103 		if (mnt->mnt.mnt_flags & MNT_NOSUID)
1104 			mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID;
1105 
1106 		if (mnt->mnt.mnt_flags & MNT_NOEXEC)
1107 			mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC;
1108 	}
1109 
1110 	/* Don't allow unprivileged users to reveal what is under a mount */
1111 	if ((flag & CL_UNPRIVILEGED) &&
1112 	    (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire)))
1113 		mnt->mnt.mnt_flags |= MNT_LOCKED;
1114 
1115 	atomic_inc(&sb->s_active);
1116 	mnt->mnt.mnt_sb = sb;
1117 	mnt->mnt.mnt_root = dget(root);
1118 	mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1119 	mnt->mnt_parent = mnt;
1120 	lock_mount_hash();
1121 	list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1122 	unlock_mount_hash();
1123 
1124 	if ((flag & CL_SLAVE) ||
1125 	    ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1126 		list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1127 		mnt->mnt_master = old;
1128 		CLEAR_MNT_SHARED(mnt);
1129 	} else if (!(flag & CL_PRIVATE)) {
1130 		if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1131 			list_add(&mnt->mnt_share, &old->mnt_share);
1132 		if (IS_MNT_SLAVE(old))
1133 			list_add(&mnt->mnt_slave, &old->mnt_slave);
1134 		mnt->mnt_master = old->mnt_master;
1135 	} else {
1136 		CLEAR_MNT_SHARED(mnt);
1137 	}
1138 	if (flag & CL_MAKE_SHARED)
1139 		set_mnt_shared(mnt);
1140 
1141 	/* stick the duplicate mount on the same expiry list
1142 	 * as the original if that was on one */
1143 	if (flag & CL_EXPIRE) {
1144 		if (!list_empty(&old->mnt_expire))
1145 			list_add(&mnt->mnt_expire, &old->mnt_expire);
1146 	}
1147 
1148 	return mnt;
1149 
1150  out_free:
1151 	mnt_free_id(mnt);
1152 	free_vfsmnt(mnt);
1153 	return ERR_PTR(err);
1154 }
1155 
1156 static void cleanup_mnt(struct mount *mnt)
1157 {
1158 	/*
1159 	 * This probably indicates that somebody messed
1160 	 * up a mnt_want/drop_write() pair.  If this
1161 	 * happens, the filesystem was probably unable
1162 	 * to make r/w->r/o transitions.
1163 	 */
1164 	/*
1165 	 * The locking used to deal with mnt_count decrement provides barriers,
1166 	 * so mnt_get_writers() below is safe.
1167 	 */
1168 	WARN_ON(mnt_get_writers(mnt));
1169 	if (unlikely(mnt->mnt_pins.first))
1170 		mnt_pin_kill(mnt);
1171 	fsnotify_vfsmount_delete(&mnt->mnt);
1172 	dput(mnt->mnt.mnt_root);
1173 	deactivate_super(mnt->mnt.mnt_sb);
1174 	mnt_free_id(mnt);
1175 	call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1176 }
1177 
1178 static void __cleanup_mnt(struct rcu_head *head)
1179 {
1180 	cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1181 }
1182 
1183 static LLIST_HEAD(delayed_mntput_list);
1184 static void delayed_mntput(struct work_struct *unused)
1185 {
1186 	struct llist_node *node = llist_del_all(&delayed_mntput_list);
1187 	struct mount *m, *t;
1188 
1189 	llist_for_each_entry_safe(m, t, node, mnt_llist)
1190 		cleanup_mnt(m);
1191 }
1192 static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1193 
1194 static void mntput_no_expire(struct mount *mnt)
1195 {
1196 	rcu_read_lock();
1197 	mnt_add_count(mnt, -1);
1198 	if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
1199 		rcu_read_unlock();
1200 		return;
1201 	}
1202 	lock_mount_hash();
1203 	if (mnt_get_count(mnt)) {
1204 		rcu_read_unlock();
1205 		unlock_mount_hash();
1206 		return;
1207 	}
1208 	if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1209 		rcu_read_unlock();
1210 		unlock_mount_hash();
1211 		return;
1212 	}
1213 	mnt->mnt.mnt_flags |= MNT_DOOMED;
1214 	rcu_read_unlock();
1215 
1216 	list_del(&mnt->mnt_instance);
1217 
1218 	if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1219 		struct mount *p, *tmp;
1220 		list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1221 			umount_mnt(p);
1222 		}
1223 	}
1224 	unlock_mount_hash();
1225 
1226 	if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1227 		struct task_struct *task = current;
1228 		if (likely(!(task->flags & PF_KTHREAD))) {
1229 			init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1230 			if (!task_work_add(task, &mnt->mnt_rcu, true))
1231 				return;
1232 		}
1233 		if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1234 			schedule_delayed_work(&delayed_mntput_work, 1);
1235 		return;
1236 	}
1237 	cleanup_mnt(mnt);
1238 }
1239 
1240 void mntput(struct vfsmount *mnt)
1241 {
1242 	if (mnt) {
1243 		struct mount *m = real_mount(mnt);
1244 		/* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1245 		if (unlikely(m->mnt_expiry_mark))
1246 			m->mnt_expiry_mark = 0;
1247 		mntput_no_expire(m);
1248 	}
1249 }
1250 EXPORT_SYMBOL(mntput);
1251 
1252 struct vfsmount *mntget(struct vfsmount *mnt)
1253 {
1254 	if (mnt)
1255 		mnt_add_count(real_mount(mnt), 1);
1256 	return mnt;
1257 }
1258 EXPORT_SYMBOL(mntget);
1259 
1260 /* path_is_mountpoint() - Check if path is a mount in the current
1261  *                          namespace.
1262  *
1263  *  d_mountpoint() can only be used reliably to establish if a dentry is
1264  *  not mounted in any namespace and that common case is handled inline.
1265  *  d_mountpoint() isn't aware of the possibility there may be multiple
1266  *  mounts using a given dentry in a different namespace. This function
1267  *  checks if the passed in path is a mountpoint rather than the dentry
1268  *  alone.
1269  */
1270 bool path_is_mountpoint(const struct path *path)
1271 {
1272 	unsigned seq;
1273 	bool res;
1274 
1275 	if (!d_mountpoint(path->dentry))
1276 		return false;
1277 
1278 	rcu_read_lock();
1279 	do {
1280 		seq = read_seqbegin(&mount_lock);
1281 		res = __path_is_mountpoint(path);
1282 	} while (read_seqretry(&mount_lock, seq));
1283 	rcu_read_unlock();
1284 
1285 	return res;
1286 }
1287 EXPORT_SYMBOL(path_is_mountpoint);
1288 
1289 struct vfsmount *mnt_clone_internal(const struct path *path)
1290 {
1291 	struct mount *p;
1292 	p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1293 	if (IS_ERR(p))
1294 		return ERR_CAST(p);
1295 	p->mnt.mnt_flags |= MNT_INTERNAL;
1296 	return &p->mnt;
1297 }
1298 
1299 #ifdef CONFIG_PROC_FS
1300 /* iterator; we want it to have access to namespace_sem, thus here... */
1301 static void *m_start(struct seq_file *m, loff_t *pos)
1302 {
1303 	struct proc_mounts *p = m->private;
1304 
1305 	down_read(&namespace_sem);
1306 	if (p->cached_event == p->ns->event) {
1307 		void *v = p->cached_mount;
1308 		if (*pos == p->cached_index)
1309 			return v;
1310 		if (*pos == p->cached_index + 1) {
1311 			v = seq_list_next(v, &p->ns->list, &p->cached_index);
1312 			return p->cached_mount = v;
1313 		}
1314 	}
1315 
1316 	p->cached_event = p->ns->event;
1317 	p->cached_mount = seq_list_start(&p->ns->list, *pos);
1318 	p->cached_index = *pos;
1319 	return p->cached_mount;
1320 }
1321 
1322 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1323 {
1324 	struct proc_mounts *p = m->private;
1325 
1326 	p->cached_mount = seq_list_next(v, &p->ns->list, pos);
1327 	p->cached_index = *pos;
1328 	return p->cached_mount;
1329 }
1330 
1331 static void m_stop(struct seq_file *m, void *v)
1332 {
1333 	up_read(&namespace_sem);
1334 }
1335 
1336 static int m_show(struct seq_file *m, void *v)
1337 {
1338 	struct proc_mounts *p = m->private;
1339 	struct mount *r = list_entry(v, struct mount, mnt_list);
1340 	return p->show(m, &r->mnt);
1341 }
1342 
1343 const struct seq_operations mounts_op = {
1344 	.start	= m_start,
1345 	.next	= m_next,
1346 	.stop	= m_stop,
1347 	.show	= m_show,
1348 };
1349 #endif  /* CONFIG_PROC_FS */
1350 
1351 /**
1352  * may_umount_tree - check if a mount tree is busy
1353  * @mnt: root of mount tree
1354  *
1355  * This is called to check if a tree of mounts has any
1356  * open files, pwds, chroots or sub mounts that are
1357  * busy.
1358  */
1359 int may_umount_tree(struct vfsmount *m)
1360 {
1361 	struct mount *mnt = real_mount(m);
1362 	int actual_refs = 0;
1363 	int minimum_refs = 0;
1364 	struct mount *p;
1365 	BUG_ON(!m);
1366 
1367 	/* write lock needed for mnt_get_count */
1368 	lock_mount_hash();
1369 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1370 		actual_refs += mnt_get_count(p);
1371 		minimum_refs += 2;
1372 	}
1373 	unlock_mount_hash();
1374 
1375 	if (actual_refs > minimum_refs)
1376 		return 0;
1377 
1378 	return 1;
1379 }
1380 
1381 EXPORT_SYMBOL(may_umount_tree);
1382 
1383 /**
1384  * may_umount - check if a mount point is busy
1385  * @mnt: root of mount
1386  *
1387  * This is called to check if a mount point has any
1388  * open files, pwds, chroots or sub mounts. If the
1389  * mount has sub mounts this will return busy
1390  * regardless of whether the sub mounts are busy.
1391  *
1392  * Doesn't take quota and stuff into account. IOW, in some cases it will
1393  * give false negatives. The main reason why it's here is that we need
1394  * a non-destructive way to look for easily umountable filesystems.
1395  */
1396 int may_umount(struct vfsmount *mnt)
1397 {
1398 	int ret = 1;
1399 	down_read(&namespace_sem);
1400 	lock_mount_hash();
1401 	if (propagate_mount_busy(real_mount(mnt), 2))
1402 		ret = 0;
1403 	unlock_mount_hash();
1404 	up_read(&namespace_sem);
1405 	return ret;
1406 }
1407 
1408 EXPORT_SYMBOL(may_umount);
1409 
1410 static HLIST_HEAD(unmounted);	/* protected by namespace_sem */
1411 
1412 static void namespace_unlock(void)
1413 {
1414 	struct hlist_head head;
1415 
1416 	hlist_move_list(&unmounted, &head);
1417 
1418 	up_write(&namespace_sem);
1419 
1420 	if (likely(hlist_empty(&head)))
1421 		return;
1422 
1423 	synchronize_rcu();
1424 
1425 	group_pin_kill(&head);
1426 }
1427 
1428 static inline void namespace_lock(void)
1429 {
1430 	down_write(&namespace_sem);
1431 }
1432 
1433 enum umount_tree_flags {
1434 	UMOUNT_SYNC = 1,
1435 	UMOUNT_PROPAGATE = 2,
1436 	UMOUNT_CONNECTED = 4,
1437 };
1438 
1439 static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1440 {
1441 	/* Leaving mounts connected is only valid for lazy umounts */
1442 	if (how & UMOUNT_SYNC)
1443 		return true;
1444 
1445 	/* A mount without a parent has nothing to be connected to */
1446 	if (!mnt_has_parent(mnt))
1447 		return true;
1448 
1449 	/* Because the reference counting rules change when mounts are
1450 	 * unmounted and connected, umounted mounts may not be
1451 	 * connected to mounted mounts.
1452 	 */
1453 	if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1454 		return true;
1455 
1456 	/* Has it been requested that the mount remain connected? */
1457 	if (how & UMOUNT_CONNECTED)
1458 		return false;
1459 
1460 	/* Is the mount locked such that it needs to remain connected? */
1461 	if (IS_MNT_LOCKED(mnt))
1462 		return false;
1463 
1464 	/* By default disconnect the mount */
1465 	return true;
1466 }
1467 
1468 /*
1469  * mount_lock must be held
1470  * namespace_sem must be held for write
1471  */
1472 static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1473 {
1474 	LIST_HEAD(tmp_list);
1475 	struct mount *p;
1476 
1477 	if (how & UMOUNT_PROPAGATE)
1478 		propagate_mount_unlock(mnt);
1479 
1480 	/* Gather the mounts to umount */
1481 	for (p = mnt; p; p = next_mnt(p, mnt)) {
1482 		p->mnt.mnt_flags |= MNT_UMOUNT;
1483 		list_move(&p->mnt_list, &tmp_list);
1484 	}
1485 
1486 	/* Hide the mounts from mnt_mounts */
1487 	list_for_each_entry(p, &tmp_list, mnt_list) {
1488 		list_del_init(&p->mnt_child);
1489 	}
1490 
1491 	/* Add propogated mounts to the tmp_list */
1492 	if (how & UMOUNT_PROPAGATE)
1493 		propagate_umount(&tmp_list);
1494 
1495 	while (!list_empty(&tmp_list)) {
1496 		struct mnt_namespace *ns;
1497 		bool disconnect;
1498 		p = list_first_entry(&tmp_list, struct mount, mnt_list);
1499 		list_del_init(&p->mnt_expire);
1500 		list_del_init(&p->mnt_list);
1501 		ns = p->mnt_ns;
1502 		if (ns) {
1503 			ns->mounts--;
1504 			__touch_mnt_namespace(ns);
1505 		}
1506 		p->mnt_ns = NULL;
1507 		if (how & UMOUNT_SYNC)
1508 			p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1509 
1510 		disconnect = disconnect_mount(p, how);
1511 
1512 		pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt,
1513 				 disconnect ? &unmounted : NULL);
1514 		if (mnt_has_parent(p)) {
1515 			mnt_add_count(p->mnt_parent, -1);
1516 			if (!disconnect) {
1517 				/* Don't forget about p */
1518 				list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1519 			} else {
1520 				umount_mnt(p);
1521 			}
1522 		}
1523 		change_mnt_propagation(p, MS_PRIVATE);
1524 	}
1525 }
1526 
1527 static void shrink_submounts(struct mount *mnt);
1528 
1529 static int do_umount(struct mount *mnt, int flags)
1530 {
1531 	struct super_block *sb = mnt->mnt.mnt_sb;
1532 	int retval;
1533 
1534 	retval = security_sb_umount(&mnt->mnt, flags);
1535 	if (retval)
1536 		return retval;
1537 
1538 	/*
1539 	 * Allow userspace to request a mountpoint be expired rather than
1540 	 * unmounting unconditionally. Unmount only happens if:
1541 	 *  (1) the mark is already set (the mark is cleared by mntput())
1542 	 *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1543 	 */
1544 	if (flags & MNT_EXPIRE) {
1545 		if (&mnt->mnt == current->fs->root.mnt ||
1546 		    flags & (MNT_FORCE | MNT_DETACH))
1547 			return -EINVAL;
1548 
1549 		/*
1550 		 * probably don't strictly need the lock here if we examined
1551 		 * all race cases, but it's a slowpath.
1552 		 */
1553 		lock_mount_hash();
1554 		if (mnt_get_count(mnt) != 2) {
1555 			unlock_mount_hash();
1556 			return -EBUSY;
1557 		}
1558 		unlock_mount_hash();
1559 
1560 		if (!xchg(&mnt->mnt_expiry_mark, 1))
1561 			return -EAGAIN;
1562 	}
1563 
1564 	/*
1565 	 * If we may have to abort operations to get out of this
1566 	 * mount, and they will themselves hold resources we must
1567 	 * allow the fs to do things. In the Unix tradition of
1568 	 * 'Gee thats tricky lets do it in userspace' the umount_begin
1569 	 * might fail to complete on the first run through as other tasks
1570 	 * must return, and the like. Thats for the mount program to worry
1571 	 * about for the moment.
1572 	 */
1573 
1574 	if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1575 		sb->s_op->umount_begin(sb);
1576 	}
1577 
1578 	/*
1579 	 * No sense to grab the lock for this test, but test itself looks
1580 	 * somewhat bogus. Suggestions for better replacement?
1581 	 * Ho-hum... In principle, we might treat that as umount + switch
1582 	 * to rootfs. GC would eventually take care of the old vfsmount.
1583 	 * Actually it makes sense, especially if rootfs would contain a
1584 	 * /reboot - static binary that would close all descriptors and
1585 	 * call reboot(9). Then init(8) could umount root and exec /reboot.
1586 	 */
1587 	if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1588 		/*
1589 		 * Special case for "unmounting" root ...
1590 		 * we just try to remount it readonly.
1591 		 */
1592 		if (!capable(CAP_SYS_ADMIN))
1593 			return -EPERM;
1594 		down_write(&sb->s_umount);
1595 		if (!sb_rdonly(sb))
1596 			retval = do_remount_sb(sb, SB_RDONLY, NULL, 0);
1597 		up_write(&sb->s_umount);
1598 		return retval;
1599 	}
1600 
1601 	namespace_lock();
1602 	lock_mount_hash();
1603 	event++;
1604 
1605 	if (flags & MNT_DETACH) {
1606 		if (!list_empty(&mnt->mnt_list))
1607 			umount_tree(mnt, UMOUNT_PROPAGATE);
1608 		retval = 0;
1609 	} else {
1610 		shrink_submounts(mnt);
1611 		retval = -EBUSY;
1612 		if (!propagate_mount_busy(mnt, 2)) {
1613 			if (!list_empty(&mnt->mnt_list))
1614 				umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1615 			retval = 0;
1616 		}
1617 	}
1618 	unlock_mount_hash();
1619 	namespace_unlock();
1620 	return retval;
1621 }
1622 
1623 /*
1624  * __detach_mounts - lazily unmount all mounts on the specified dentry
1625  *
1626  * During unlink, rmdir, and d_drop it is possible to loose the path
1627  * to an existing mountpoint, and wind up leaking the mount.
1628  * detach_mounts allows lazily unmounting those mounts instead of
1629  * leaking them.
1630  *
1631  * The caller may hold dentry->d_inode->i_mutex.
1632  */
1633 void __detach_mounts(struct dentry *dentry)
1634 {
1635 	struct mountpoint *mp;
1636 	struct mount *mnt;
1637 
1638 	namespace_lock();
1639 	lock_mount_hash();
1640 	mp = lookup_mountpoint(dentry);
1641 	if (IS_ERR_OR_NULL(mp))
1642 		goto out_unlock;
1643 
1644 	event++;
1645 	while (!hlist_empty(&mp->m_list)) {
1646 		mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1647 		if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1648 			hlist_add_head(&mnt->mnt_umount.s_list, &unmounted);
1649 			umount_mnt(mnt);
1650 		}
1651 		else umount_tree(mnt, UMOUNT_CONNECTED);
1652 	}
1653 	put_mountpoint(mp);
1654 out_unlock:
1655 	unlock_mount_hash();
1656 	namespace_unlock();
1657 }
1658 
1659 /*
1660  * Is the caller allowed to modify his namespace?
1661  */
1662 static inline bool may_mount(void)
1663 {
1664 	return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1665 }
1666 
1667 static inline bool may_mandlock(void)
1668 {
1669 #ifndef	CONFIG_MANDATORY_FILE_LOCKING
1670 	return false;
1671 #endif
1672 	return capable(CAP_SYS_ADMIN);
1673 }
1674 
1675 /*
1676  * Now umount can handle mount points as well as block devices.
1677  * This is important for filesystems which use unnamed block devices.
1678  *
1679  * We now support a flag for forced unmount like the other 'big iron'
1680  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1681  */
1682 
1683 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1684 {
1685 	struct path path;
1686 	struct mount *mnt;
1687 	int retval;
1688 	int lookup_flags = 0;
1689 
1690 	if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1691 		return -EINVAL;
1692 
1693 	if (!may_mount())
1694 		return -EPERM;
1695 
1696 	if (!(flags & UMOUNT_NOFOLLOW))
1697 		lookup_flags |= LOOKUP_FOLLOW;
1698 
1699 	retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1700 	if (retval)
1701 		goto out;
1702 	mnt = real_mount(path.mnt);
1703 	retval = -EINVAL;
1704 	if (path.dentry != path.mnt->mnt_root)
1705 		goto dput_and_out;
1706 	if (!check_mnt(mnt))
1707 		goto dput_and_out;
1708 	if (mnt->mnt.mnt_flags & MNT_LOCKED)
1709 		goto dput_and_out;
1710 	retval = -EPERM;
1711 	if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1712 		goto dput_and_out;
1713 
1714 	retval = do_umount(mnt, flags);
1715 dput_and_out:
1716 	/* we mustn't call path_put() as that would clear mnt_expiry_mark */
1717 	dput(path.dentry);
1718 	mntput_no_expire(mnt);
1719 out:
1720 	return retval;
1721 }
1722 
1723 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1724 
1725 /*
1726  *	The 2.0 compatible umount. No flags.
1727  */
1728 SYSCALL_DEFINE1(oldumount, char __user *, name)
1729 {
1730 	return sys_umount(name, 0);
1731 }
1732 
1733 #endif
1734 
1735 static bool is_mnt_ns_file(struct dentry *dentry)
1736 {
1737 	/* Is this a proxy for a mount namespace? */
1738 	return dentry->d_op == &ns_dentry_operations &&
1739 	       dentry->d_fsdata == &mntns_operations;
1740 }
1741 
1742 struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1743 {
1744 	return container_of(ns, struct mnt_namespace, ns);
1745 }
1746 
1747 static bool mnt_ns_loop(struct dentry *dentry)
1748 {
1749 	/* Could bind mounting the mount namespace inode cause a
1750 	 * mount namespace loop?
1751 	 */
1752 	struct mnt_namespace *mnt_ns;
1753 	if (!is_mnt_ns_file(dentry))
1754 		return false;
1755 
1756 	mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1757 	return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1758 }
1759 
1760 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1761 					int flag)
1762 {
1763 	struct mount *res, *p, *q, *r, *parent;
1764 
1765 	if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1766 		return ERR_PTR(-EINVAL);
1767 
1768 	if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1769 		return ERR_PTR(-EINVAL);
1770 
1771 	res = q = clone_mnt(mnt, dentry, flag);
1772 	if (IS_ERR(q))
1773 		return q;
1774 
1775 	q->mnt_mountpoint = mnt->mnt_mountpoint;
1776 
1777 	p = mnt;
1778 	list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1779 		struct mount *s;
1780 		if (!is_subdir(r->mnt_mountpoint, dentry))
1781 			continue;
1782 
1783 		for (s = r; s; s = next_mnt(s, r)) {
1784 			if (!(flag & CL_COPY_UNBINDABLE) &&
1785 			    IS_MNT_UNBINDABLE(s)) {
1786 				s = skip_mnt_tree(s);
1787 				continue;
1788 			}
1789 			if (!(flag & CL_COPY_MNT_NS_FILE) &&
1790 			    is_mnt_ns_file(s->mnt.mnt_root)) {
1791 				s = skip_mnt_tree(s);
1792 				continue;
1793 			}
1794 			while (p != s->mnt_parent) {
1795 				p = p->mnt_parent;
1796 				q = q->mnt_parent;
1797 			}
1798 			p = s;
1799 			parent = q;
1800 			q = clone_mnt(p, p->mnt.mnt_root, flag);
1801 			if (IS_ERR(q))
1802 				goto out;
1803 			lock_mount_hash();
1804 			list_add_tail(&q->mnt_list, &res->mnt_list);
1805 			attach_mnt(q, parent, p->mnt_mp);
1806 			unlock_mount_hash();
1807 		}
1808 	}
1809 	return res;
1810 out:
1811 	if (res) {
1812 		lock_mount_hash();
1813 		umount_tree(res, UMOUNT_SYNC);
1814 		unlock_mount_hash();
1815 	}
1816 	return q;
1817 }
1818 
1819 /* Caller should check returned pointer for errors */
1820 
1821 struct vfsmount *collect_mounts(const struct path *path)
1822 {
1823 	struct mount *tree;
1824 	namespace_lock();
1825 	if (!check_mnt(real_mount(path->mnt)))
1826 		tree = ERR_PTR(-EINVAL);
1827 	else
1828 		tree = copy_tree(real_mount(path->mnt), path->dentry,
1829 				 CL_COPY_ALL | CL_PRIVATE);
1830 	namespace_unlock();
1831 	if (IS_ERR(tree))
1832 		return ERR_CAST(tree);
1833 	return &tree->mnt;
1834 }
1835 
1836 void drop_collected_mounts(struct vfsmount *mnt)
1837 {
1838 	namespace_lock();
1839 	lock_mount_hash();
1840 	umount_tree(real_mount(mnt), UMOUNT_SYNC);
1841 	unlock_mount_hash();
1842 	namespace_unlock();
1843 }
1844 
1845 /**
1846  * clone_private_mount - create a private clone of a path
1847  *
1848  * This creates a new vfsmount, which will be the clone of @path.  The new will
1849  * not be attached anywhere in the namespace and will be private (i.e. changes
1850  * to the originating mount won't be propagated into this).
1851  *
1852  * Release with mntput().
1853  */
1854 struct vfsmount *clone_private_mount(const struct path *path)
1855 {
1856 	struct mount *old_mnt = real_mount(path->mnt);
1857 	struct mount *new_mnt;
1858 
1859 	if (IS_MNT_UNBINDABLE(old_mnt))
1860 		return ERR_PTR(-EINVAL);
1861 
1862 	new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1863 	if (IS_ERR(new_mnt))
1864 		return ERR_CAST(new_mnt);
1865 
1866 	return &new_mnt->mnt;
1867 }
1868 EXPORT_SYMBOL_GPL(clone_private_mount);
1869 
1870 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1871 		   struct vfsmount *root)
1872 {
1873 	struct mount *mnt;
1874 	int res = f(root, arg);
1875 	if (res)
1876 		return res;
1877 	list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1878 		res = f(&mnt->mnt, arg);
1879 		if (res)
1880 			return res;
1881 	}
1882 	return 0;
1883 }
1884 
1885 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1886 {
1887 	struct mount *p;
1888 
1889 	for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1890 		if (p->mnt_group_id && !IS_MNT_SHARED(p))
1891 			mnt_release_group_id(p);
1892 	}
1893 }
1894 
1895 static int invent_group_ids(struct mount *mnt, bool recurse)
1896 {
1897 	struct mount *p;
1898 
1899 	for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1900 		if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1901 			int err = mnt_alloc_group_id(p);
1902 			if (err) {
1903 				cleanup_group_ids(mnt, p);
1904 				return err;
1905 			}
1906 		}
1907 	}
1908 
1909 	return 0;
1910 }
1911 
1912 int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
1913 {
1914 	unsigned int max = READ_ONCE(sysctl_mount_max);
1915 	unsigned int mounts = 0, old, pending, sum;
1916 	struct mount *p;
1917 
1918 	for (p = mnt; p; p = next_mnt(p, mnt))
1919 		mounts++;
1920 
1921 	old = ns->mounts;
1922 	pending = ns->pending_mounts;
1923 	sum = old + pending;
1924 	if ((old > sum) ||
1925 	    (pending > sum) ||
1926 	    (max < sum) ||
1927 	    (mounts > (max - sum)))
1928 		return -ENOSPC;
1929 
1930 	ns->pending_mounts = pending + mounts;
1931 	return 0;
1932 }
1933 
1934 /*
1935  *  @source_mnt : mount tree to be attached
1936  *  @nd         : place the mount tree @source_mnt is attached
1937  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1938  *  		   store the parent mount and mountpoint dentry.
1939  *  		   (done when source_mnt is moved)
1940  *
1941  *  NOTE: in the table below explains the semantics when a source mount
1942  *  of a given type is attached to a destination mount of a given type.
1943  * ---------------------------------------------------------------------------
1944  * |         BIND MOUNT OPERATION                                            |
1945  * |**************************************************************************
1946  * | source-->| shared        |       private  |       slave    | unbindable |
1947  * | dest     |               |                |                |            |
1948  * |   |      |               |                |                |            |
1949  * |   v      |               |                |                |            |
1950  * |**************************************************************************
1951  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1952  * |          |               |                |                |            |
1953  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1954  * ***************************************************************************
1955  * A bind operation clones the source mount and mounts the clone on the
1956  * destination mount.
1957  *
1958  * (++)  the cloned mount is propagated to all the mounts in the propagation
1959  * 	 tree of the destination mount and the cloned mount is added to
1960  * 	 the peer group of the source mount.
1961  * (+)   the cloned mount is created under the destination mount and is marked
1962  *       as shared. The cloned mount is added to the peer group of the source
1963  *       mount.
1964  * (+++) the mount is propagated to all the mounts in the propagation tree
1965  *       of the destination mount and the cloned mount is made slave
1966  *       of the same master as that of the source mount. The cloned mount
1967  *       is marked as 'shared and slave'.
1968  * (*)   the cloned mount is made a slave of the same master as that of the
1969  * 	 source mount.
1970  *
1971  * ---------------------------------------------------------------------------
1972  * |         		MOVE MOUNT OPERATION                                 |
1973  * |**************************************************************************
1974  * | source-->| shared        |       private  |       slave    | unbindable |
1975  * | dest     |               |                |                |            |
1976  * |   |      |               |                |                |            |
1977  * |   v      |               |                |                |            |
1978  * |**************************************************************************
1979  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1980  * |          |               |                |                |            |
1981  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1982  * ***************************************************************************
1983  *
1984  * (+)  the mount is moved to the destination. And is then propagated to
1985  * 	all the mounts in the propagation tree of the destination mount.
1986  * (+*)  the mount is moved to the destination.
1987  * (+++)  the mount is moved to the destination and is then propagated to
1988  * 	all the mounts belonging to the destination mount's propagation tree.
1989  * 	the mount is marked as 'shared and slave'.
1990  * (*)	the mount continues to be a slave at the new location.
1991  *
1992  * if the source mount is a tree, the operations explained above is
1993  * applied to each mount in the tree.
1994  * Must be called without spinlocks held, since this function can sleep
1995  * in allocations.
1996  */
1997 static int attach_recursive_mnt(struct mount *source_mnt,
1998 			struct mount *dest_mnt,
1999 			struct mountpoint *dest_mp,
2000 			struct path *parent_path)
2001 {
2002 	HLIST_HEAD(tree_list);
2003 	struct mnt_namespace *ns = dest_mnt->mnt_ns;
2004 	struct mountpoint *smp;
2005 	struct mount *child, *p;
2006 	struct hlist_node *n;
2007 	int err;
2008 
2009 	/* Preallocate a mountpoint in case the new mounts need
2010 	 * to be tucked under other mounts.
2011 	 */
2012 	smp = get_mountpoint(source_mnt->mnt.mnt_root);
2013 	if (IS_ERR(smp))
2014 		return PTR_ERR(smp);
2015 
2016 	/* Is there space to add these mounts to the mount namespace? */
2017 	if (!parent_path) {
2018 		err = count_mounts(ns, source_mnt);
2019 		if (err)
2020 			goto out;
2021 	}
2022 
2023 	if (IS_MNT_SHARED(dest_mnt)) {
2024 		err = invent_group_ids(source_mnt, true);
2025 		if (err)
2026 			goto out;
2027 		err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2028 		lock_mount_hash();
2029 		if (err)
2030 			goto out_cleanup_ids;
2031 		for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2032 			set_mnt_shared(p);
2033 	} else {
2034 		lock_mount_hash();
2035 	}
2036 	if (parent_path) {
2037 		detach_mnt(source_mnt, parent_path);
2038 		attach_mnt(source_mnt, dest_mnt, dest_mp);
2039 		touch_mnt_namespace(source_mnt->mnt_ns);
2040 	} else {
2041 		mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2042 		commit_tree(source_mnt);
2043 	}
2044 
2045 	hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2046 		struct mount *q;
2047 		hlist_del_init(&child->mnt_hash);
2048 		q = __lookup_mnt(&child->mnt_parent->mnt,
2049 				 child->mnt_mountpoint);
2050 		if (q)
2051 			mnt_change_mountpoint(child, smp, q);
2052 		commit_tree(child);
2053 	}
2054 	put_mountpoint(smp);
2055 	unlock_mount_hash();
2056 
2057 	return 0;
2058 
2059  out_cleanup_ids:
2060 	while (!hlist_empty(&tree_list)) {
2061 		child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2062 		child->mnt_parent->mnt_ns->pending_mounts = 0;
2063 		umount_tree(child, UMOUNT_SYNC);
2064 	}
2065 	unlock_mount_hash();
2066 	cleanup_group_ids(source_mnt, NULL);
2067  out:
2068 	ns->pending_mounts = 0;
2069 
2070 	read_seqlock_excl(&mount_lock);
2071 	put_mountpoint(smp);
2072 	read_sequnlock_excl(&mount_lock);
2073 
2074 	return err;
2075 }
2076 
2077 static struct mountpoint *lock_mount(struct path *path)
2078 {
2079 	struct vfsmount *mnt;
2080 	struct dentry *dentry = path->dentry;
2081 retry:
2082 	inode_lock(dentry->d_inode);
2083 	if (unlikely(cant_mount(dentry))) {
2084 		inode_unlock(dentry->d_inode);
2085 		return ERR_PTR(-ENOENT);
2086 	}
2087 	namespace_lock();
2088 	mnt = lookup_mnt(path);
2089 	if (likely(!mnt)) {
2090 		struct mountpoint *mp = get_mountpoint(dentry);
2091 		if (IS_ERR(mp)) {
2092 			namespace_unlock();
2093 			inode_unlock(dentry->d_inode);
2094 			return mp;
2095 		}
2096 		return mp;
2097 	}
2098 	namespace_unlock();
2099 	inode_unlock(path->dentry->d_inode);
2100 	path_put(path);
2101 	path->mnt = mnt;
2102 	dentry = path->dentry = dget(mnt->mnt_root);
2103 	goto retry;
2104 }
2105 
2106 static void unlock_mount(struct mountpoint *where)
2107 {
2108 	struct dentry *dentry = where->m_dentry;
2109 
2110 	read_seqlock_excl(&mount_lock);
2111 	put_mountpoint(where);
2112 	read_sequnlock_excl(&mount_lock);
2113 
2114 	namespace_unlock();
2115 	inode_unlock(dentry->d_inode);
2116 }
2117 
2118 static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2119 {
2120 	if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2121 		return -EINVAL;
2122 
2123 	if (d_is_dir(mp->m_dentry) !=
2124 	      d_is_dir(mnt->mnt.mnt_root))
2125 		return -ENOTDIR;
2126 
2127 	return attach_recursive_mnt(mnt, p, mp, NULL);
2128 }
2129 
2130 /*
2131  * Sanity check the flags to change_mnt_propagation.
2132  */
2133 
2134 static int flags_to_propagation_type(int ms_flags)
2135 {
2136 	int type = ms_flags & ~(MS_REC | MS_SILENT);
2137 
2138 	/* Fail if any non-propagation flags are set */
2139 	if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2140 		return 0;
2141 	/* Only one propagation flag should be set */
2142 	if (!is_power_of_2(type))
2143 		return 0;
2144 	return type;
2145 }
2146 
2147 /*
2148  * recursively change the type of the mountpoint.
2149  */
2150 static int do_change_type(struct path *path, int ms_flags)
2151 {
2152 	struct mount *m;
2153 	struct mount *mnt = real_mount(path->mnt);
2154 	int recurse = ms_flags & MS_REC;
2155 	int type;
2156 	int err = 0;
2157 
2158 	if (path->dentry != path->mnt->mnt_root)
2159 		return -EINVAL;
2160 
2161 	type = flags_to_propagation_type(ms_flags);
2162 	if (!type)
2163 		return -EINVAL;
2164 
2165 	namespace_lock();
2166 	if (type == MS_SHARED) {
2167 		err = invent_group_ids(mnt, recurse);
2168 		if (err)
2169 			goto out_unlock;
2170 	}
2171 
2172 	lock_mount_hash();
2173 	for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2174 		change_mnt_propagation(m, type);
2175 	unlock_mount_hash();
2176 
2177  out_unlock:
2178 	namespace_unlock();
2179 	return err;
2180 }
2181 
2182 static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2183 {
2184 	struct mount *child;
2185 	list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2186 		if (!is_subdir(child->mnt_mountpoint, dentry))
2187 			continue;
2188 
2189 		if (child->mnt.mnt_flags & MNT_LOCKED)
2190 			return true;
2191 	}
2192 	return false;
2193 }
2194 
2195 /*
2196  * do loopback mount.
2197  */
2198 static int do_loopback(struct path *path, const char *old_name,
2199 				int recurse)
2200 {
2201 	struct path old_path;
2202 	struct mount *mnt = NULL, *old, *parent;
2203 	struct mountpoint *mp;
2204 	int err;
2205 	if (!old_name || !*old_name)
2206 		return -EINVAL;
2207 	err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2208 	if (err)
2209 		return err;
2210 
2211 	err = -EINVAL;
2212 	if (mnt_ns_loop(old_path.dentry))
2213 		goto out;
2214 
2215 	mp = lock_mount(path);
2216 	err = PTR_ERR(mp);
2217 	if (IS_ERR(mp))
2218 		goto out;
2219 
2220 	old = real_mount(old_path.mnt);
2221 	parent = real_mount(path->mnt);
2222 
2223 	err = -EINVAL;
2224 	if (IS_MNT_UNBINDABLE(old))
2225 		goto out2;
2226 
2227 	if (!check_mnt(parent))
2228 		goto out2;
2229 
2230 	if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations)
2231 		goto out2;
2232 
2233 	if (!recurse && has_locked_children(old, old_path.dentry))
2234 		goto out2;
2235 
2236 	if (recurse)
2237 		mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
2238 	else
2239 		mnt = clone_mnt(old, old_path.dentry, 0);
2240 
2241 	if (IS_ERR(mnt)) {
2242 		err = PTR_ERR(mnt);
2243 		goto out2;
2244 	}
2245 
2246 	mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2247 
2248 	err = graft_tree(mnt, parent, mp);
2249 	if (err) {
2250 		lock_mount_hash();
2251 		umount_tree(mnt, UMOUNT_SYNC);
2252 		unlock_mount_hash();
2253 	}
2254 out2:
2255 	unlock_mount(mp);
2256 out:
2257 	path_put(&old_path);
2258 	return err;
2259 }
2260 
2261 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
2262 {
2263 	int error = 0;
2264 	int readonly_request = 0;
2265 
2266 	if (ms_flags & MS_RDONLY)
2267 		readonly_request = 1;
2268 	if (readonly_request == __mnt_is_readonly(mnt))
2269 		return 0;
2270 
2271 	if (readonly_request)
2272 		error = mnt_make_readonly(real_mount(mnt));
2273 	else
2274 		__mnt_unmake_readonly(real_mount(mnt));
2275 	return error;
2276 }
2277 
2278 /*
2279  * change filesystem flags. dir should be a physical root of filesystem.
2280  * If you've mounted a non-root directory somewhere and want to do remount
2281  * on it - tough luck.
2282  */
2283 static int do_remount(struct path *path, int ms_flags, int sb_flags,
2284 		      int mnt_flags, void *data)
2285 {
2286 	int err;
2287 	struct super_block *sb = path->mnt->mnt_sb;
2288 	struct mount *mnt = real_mount(path->mnt);
2289 
2290 	if (!check_mnt(mnt))
2291 		return -EINVAL;
2292 
2293 	if (path->dentry != path->mnt->mnt_root)
2294 		return -EINVAL;
2295 
2296 	/* Don't allow changing of locked mnt flags.
2297 	 *
2298 	 * No locks need to be held here while testing the various
2299 	 * MNT_LOCK flags because those flags can never be cleared
2300 	 * once they are set.
2301 	 */
2302 	if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) &&
2303 	    !(mnt_flags & MNT_READONLY)) {
2304 		return -EPERM;
2305 	}
2306 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) &&
2307 	    !(mnt_flags & MNT_NODEV)) {
2308 		return -EPERM;
2309 	}
2310 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) &&
2311 	    !(mnt_flags & MNT_NOSUID)) {
2312 		return -EPERM;
2313 	}
2314 	if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) &&
2315 	    !(mnt_flags & MNT_NOEXEC)) {
2316 		return -EPERM;
2317 	}
2318 	if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) &&
2319 	    ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) {
2320 		return -EPERM;
2321 	}
2322 
2323 	err = security_sb_remount(sb, data);
2324 	if (err)
2325 		return err;
2326 
2327 	down_write(&sb->s_umount);
2328 	if (ms_flags & MS_BIND)
2329 		err = change_mount_flags(path->mnt, ms_flags);
2330 	else if (!capable(CAP_SYS_ADMIN))
2331 		err = -EPERM;
2332 	else
2333 		err = do_remount_sb(sb, sb_flags, data, 0);
2334 	if (!err) {
2335 		lock_mount_hash();
2336 		mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2337 		mnt->mnt.mnt_flags = mnt_flags;
2338 		touch_mnt_namespace(mnt->mnt_ns);
2339 		unlock_mount_hash();
2340 	}
2341 	up_write(&sb->s_umount);
2342 	return err;
2343 }
2344 
2345 static inline int tree_contains_unbindable(struct mount *mnt)
2346 {
2347 	struct mount *p;
2348 	for (p = mnt; p; p = next_mnt(p, mnt)) {
2349 		if (IS_MNT_UNBINDABLE(p))
2350 			return 1;
2351 	}
2352 	return 0;
2353 }
2354 
2355 static int do_move_mount(struct path *path, const char *old_name)
2356 {
2357 	struct path old_path, parent_path;
2358 	struct mount *p;
2359 	struct mount *old;
2360 	struct mountpoint *mp;
2361 	int err;
2362 	if (!old_name || !*old_name)
2363 		return -EINVAL;
2364 	err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2365 	if (err)
2366 		return err;
2367 
2368 	mp = lock_mount(path);
2369 	err = PTR_ERR(mp);
2370 	if (IS_ERR(mp))
2371 		goto out;
2372 
2373 	old = real_mount(old_path.mnt);
2374 	p = real_mount(path->mnt);
2375 
2376 	err = -EINVAL;
2377 	if (!check_mnt(p) || !check_mnt(old))
2378 		goto out1;
2379 
2380 	if (old->mnt.mnt_flags & MNT_LOCKED)
2381 		goto out1;
2382 
2383 	err = -EINVAL;
2384 	if (old_path.dentry != old_path.mnt->mnt_root)
2385 		goto out1;
2386 
2387 	if (!mnt_has_parent(old))
2388 		goto out1;
2389 
2390 	if (d_is_dir(path->dentry) !=
2391 	      d_is_dir(old_path.dentry))
2392 		goto out1;
2393 	/*
2394 	 * Don't move a mount residing in a shared parent.
2395 	 */
2396 	if (IS_MNT_SHARED(old->mnt_parent))
2397 		goto out1;
2398 	/*
2399 	 * Don't move a mount tree containing unbindable mounts to a destination
2400 	 * mount which is shared.
2401 	 */
2402 	if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2403 		goto out1;
2404 	err = -ELOOP;
2405 	for (; mnt_has_parent(p); p = p->mnt_parent)
2406 		if (p == old)
2407 			goto out1;
2408 
2409 	err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
2410 	if (err)
2411 		goto out1;
2412 
2413 	/* if the mount is moved, it should no longer be expire
2414 	 * automatically */
2415 	list_del_init(&old->mnt_expire);
2416 out1:
2417 	unlock_mount(mp);
2418 out:
2419 	if (!err)
2420 		path_put(&parent_path);
2421 	path_put(&old_path);
2422 	return err;
2423 }
2424 
2425 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
2426 {
2427 	int err;
2428 	const char *subtype = strchr(fstype, '.');
2429 	if (subtype) {
2430 		subtype++;
2431 		err = -EINVAL;
2432 		if (!subtype[0])
2433 			goto err;
2434 	} else
2435 		subtype = "";
2436 
2437 	mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
2438 	err = -ENOMEM;
2439 	if (!mnt->mnt_sb->s_subtype)
2440 		goto err;
2441 	return mnt;
2442 
2443  err:
2444 	mntput(mnt);
2445 	return ERR_PTR(err);
2446 }
2447 
2448 /*
2449  * add a mount into a namespace's mount tree
2450  */
2451 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
2452 {
2453 	struct mountpoint *mp;
2454 	struct mount *parent;
2455 	int err;
2456 
2457 	mnt_flags &= ~MNT_INTERNAL_FLAGS;
2458 
2459 	mp = lock_mount(path);
2460 	if (IS_ERR(mp))
2461 		return PTR_ERR(mp);
2462 
2463 	parent = real_mount(path->mnt);
2464 	err = -EINVAL;
2465 	if (unlikely(!check_mnt(parent))) {
2466 		/* that's acceptable only for automounts done in private ns */
2467 		if (!(mnt_flags & MNT_SHRINKABLE))
2468 			goto unlock;
2469 		/* ... and for those we'd better have mountpoint still alive */
2470 		if (!parent->mnt_ns)
2471 			goto unlock;
2472 	}
2473 
2474 	/* Refuse the same filesystem on the same mount point */
2475 	err = -EBUSY;
2476 	if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2477 	    path->mnt->mnt_root == path->dentry)
2478 		goto unlock;
2479 
2480 	err = -EINVAL;
2481 	if (d_is_symlink(newmnt->mnt.mnt_root))
2482 		goto unlock;
2483 
2484 	newmnt->mnt.mnt_flags = mnt_flags;
2485 	err = graft_tree(newmnt, parent, mp);
2486 
2487 unlock:
2488 	unlock_mount(mp);
2489 	return err;
2490 }
2491 
2492 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags);
2493 
2494 /*
2495  * create a new mount for userspace and request it to be added into the
2496  * namespace's tree
2497  */
2498 static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2499 			int mnt_flags, const char *name, void *data)
2500 {
2501 	struct file_system_type *type;
2502 	struct vfsmount *mnt;
2503 	int err;
2504 
2505 	if (!fstype)
2506 		return -EINVAL;
2507 
2508 	type = get_fs_type(fstype);
2509 	if (!type)
2510 		return -ENODEV;
2511 
2512 	mnt = vfs_kern_mount(type, sb_flags, name, data);
2513 	if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2514 	    !mnt->mnt_sb->s_subtype)
2515 		mnt = fs_set_subtype(mnt, fstype);
2516 
2517 	put_filesystem(type);
2518 	if (IS_ERR(mnt))
2519 		return PTR_ERR(mnt);
2520 
2521 	if (mount_too_revealing(mnt, &mnt_flags)) {
2522 		mntput(mnt);
2523 		return -EPERM;
2524 	}
2525 
2526 	err = do_add_mount(real_mount(mnt), path, mnt_flags);
2527 	if (err)
2528 		mntput(mnt);
2529 	return err;
2530 }
2531 
2532 int finish_automount(struct vfsmount *m, struct path *path)
2533 {
2534 	struct mount *mnt = real_mount(m);
2535 	int err;
2536 	/* The new mount record should have at least 2 refs to prevent it being
2537 	 * expired before we get a chance to add it
2538 	 */
2539 	BUG_ON(mnt_get_count(mnt) < 2);
2540 
2541 	if (m->mnt_sb == path->mnt->mnt_sb &&
2542 	    m->mnt_root == path->dentry) {
2543 		err = -ELOOP;
2544 		goto fail;
2545 	}
2546 
2547 	err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2548 	if (!err)
2549 		return 0;
2550 fail:
2551 	/* remove m from any expiration list it may be on */
2552 	if (!list_empty(&mnt->mnt_expire)) {
2553 		namespace_lock();
2554 		list_del_init(&mnt->mnt_expire);
2555 		namespace_unlock();
2556 	}
2557 	mntput(m);
2558 	mntput(m);
2559 	return err;
2560 }
2561 
2562 /**
2563  * mnt_set_expiry - Put a mount on an expiration list
2564  * @mnt: The mount to list.
2565  * @expiry_list: The list to add the mount to.
2566  */
2567 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2568 {
2569 	namespace_lock();
2570 
2571 	list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2572 
2573 	namespace_unlock();
2574 }
2575 EXPORT_SYMBOL(mnt_set_expiry);
2576 
2577 /*
2578  * process a list of expirable mountpoints with the intent of discarding any
2579  * mountpoints that aren't in use and haven't been touched since last we came
2580  * here
2581  */
2582 void mark_mounts_for_expiry(struct list_head *mounts)
2583 {
2584 	struct mount *mnt, *next;
2585 	LIST_HEAD(graveyard);
2586 
2587 	if (list_empty(mounts))
2588 		return;
2589 
2590 	namespace_lock();
2591 	lock_mount_hash();
2592 
2593 	/* extract from the expiration list every vfsmount that matches the
2594 	 * following criteria:
2595 	 * - only referenced by its parent vfsmount
2596 	 * - still marked for expiry (marked on the last call here; marks are
2597 	 *   cleared by mntput())
2598 	 */
2599 	list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2600 		if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2601 			propagate_mount_busy(mnt, 1))
2602 			continue;
2603 		list_move(&mnt->mnt_expire, &graveyard);
2604 	}
2605 	while (!list_empty(&graveyard)) {
2606 		mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2607 		touch_mnt_namespace(mnt->mnt_ns);
2608 		umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2609 	}
2610 	unlock_mount_hash();
2611 	namespace_unlock();
2612 }
2613 
2614 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2615 
2616 /*
2617  * Ripoff of 'select_parent()'
2618  *
2619  * search the list of submounts for a given mountpoint, and move any
2620  * shrinkable submounts to the 'graveyard' list.
2621  */
2622 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2623 {
2624 	struct mount *this_parent = parent;
2625 	struct list_head *next;
2626 	int found = 0;
2627 
2628 repeat:
2629 	next = this_parent->mnt_mounts.next;
2630 resume:
2631 	while (next != &this_parent->mnt_mounts) {
2632 		struct list_head *tmp = next;
2633 		struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2634 
2635 		next = tmp->next;
2636 		if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2637 			continue;
2638 		/*
2639 		 * Descend a level if the d_mounts list is non-empty.
2640 		 */
2641 		if (!list_empty(&mnt->mnt_mounts)) {
2642 			this_parent = mnt;
2643 			goto repeat;
2644 		}
2645 
2646 		if (!propagate_mount_busy(mnt, 1)) {
2647 			list_move_tail(&mnt->mnt_expire, graveyard);
2648 			found++;
2649 		}
2650 	}
2651 	/*
2652 	 * All done at this level ... ascend and resume the search
2653 	 */
2654 	if (this_parent != parent) {
2655 		next = this_parent->mnt_child.next;
2656 		this_parent = this_parent->mnt_parent;
2657 		goto resume;
2658 	}
2659 	return found;
2660 }
2661 
2662 /*
2663  * process a list of expirable mountpoints with the intent of discarding any
2664  * submounts of a specific parent mountpoint
2665  *
2666  * mount_lock must be held for write
2667  */
2668 static void shrink_submounts(struct mount *mnt)
2669 {
2670 	LIST_HEAD(graveyard);
2671 	struct mount *m;
2672 
2673 	/* extract submounts of 'mountpoint' from the expiration list */
2674 	while (select_submounts(mnt, &graveyard)) {
2675 		while (!list_empty(&graveyard)) {
2676 			m = list_first_entry(&graveyard, struct mount,
2677 						mnt_expire);
2678 			touch_mnt_namespace(m->mnt_ns);
2679 			umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
2680 		}
2681 	}
2682 }
2683 
2684 /*
2685  * Some copy_from_user() implementations do not return the exact number of
2686  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2687  * Note that this function differs from copy_from_user() in that it will oops
2688  * on bad values of `to', rather than returning a short copy.
2689  */
2690 static long exact_copy_from_user(void *to, const void __user * from,
2691 				 unsigned long n)
2692 {
2693 	char *t = to;
2694 	const char __user *f = from;
2695 	char c;
2696 
2697 	if (!access_ok(VERIFY_READ, from, n))
2698 		return n;
2699 
2700 	while (n) {
2701 		if (__get_user(c, f)) {
2702 			memset(t, 0, n);
2703 			break;
2704 		}
2705 		*t++ = c;
2706 		f++;
2707 		n--;
2708 	}
2709 	return n;
2710 }
2711 
2712 void *copy_mount_options(const void __user * data)
2713 {
2714 	int i;
2715 	unsigned long size;
2716 	char *copy;
2717 
2718 	if (!data)
2719 		return NULL;
2720 
2721 	copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
2722 	if (!copy)
2723 		return ERR_PTR(-ENOMEM);
2724 
2725 	/* We only care that *some* data at the address the user
2726 	 * gave us is valid.  Just in case, we'll zero
2727 	 * the remainder of the page.
2728 	 */
2729 	/* copy_from_user cannot cross TASK_SIZE ! */
2730 	size = TASK_SIZE - (unsigned long)data;
2731 	if (size > PAGE_SIZE)
2732 		size = PAGE_SIZE;
2733 
2734 	i = size - exact_copy_from_user(copy, data, size);
2735 	if (!i) {
2736 		kfree(copy);
2737 		return ERR_PTR(-EFAULT);
2738 	}
2739 	if (i != PAGE_SIZE)
2740 		memset(copy + i, 0, PAGE_SIZE - i);
2741 	return copy;
2742 }
2743 
2744 char *copy_mount_string(const void __user *data)
2745 {
2746 	return data ? strndup_user(data, PAGE_SIZE) : NULL;
2747 }
2748 
2749 /*
2750  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2751  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2752  *
2753  * data is a (void *) that can point to any structure up to
2754  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2755  * information (or be NULL).
2756  *
2757  * Pre-0.97 versions of mount() didn't have a flags word.
2758  * When the flags word was introduced its top half was required
2759  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2760  * Therefore, if this magic number is present, it carries no information
2761  * and must be discarded.
2762  */
2763 long do_mount(const char *dev_name, const char __user *dir_name,
2764 		const char *type_page, unsigned long flags, void *data_page)
2765 {
2766 	struct path path;
2767 	unsigned int mnt_flags = 0, sb_flags;
2768 	int retval = 0;
2769 
2770 	/* Discard magic */
2771 	if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2772 		flags &= ~MS_MGC_MSK;
2773 
2774 	/* Basic sanity checks */
2775 	if (data_page)
2776 		((char *)data_page)[PAGE_SIZE - 1] = 0;
2777 
2778 	if (flags & MS_NOUSER)
2779 		return -EINVAL;
2780 
2781 	/* ... and get the mountpoint */
2782 	retval = user_path(dir_name, &path);
2783 	if (retval)
2784 		return retval;
2785 
2786 	retval = security_sb_mount(dev_name, &path,
2787 				   type_page, flags, data_page);
2788 	if (!retval && !may_mount())
2789 		retval = -EPERM;
2790 	if (!retval && (flags & SB_MANDLOCK) && !may_mandlock())
2791 		retval = -EPERM;
2792 	if (retval)
2793 		goto dput_out;
2794 
2795 	/* Default to relatime unless overriden */
2796 	if (!(flags & MS_NOATIME))
2797 		mnt_flags |= MNT_RELATIME;
2798 
2799 	/* Separate the per-mountpoint flags */
2800 	if (flags & MS_NOSUID)
2801 		mnt_flags |= MNT_NOSUID;
2802 	if (flags & MS_NODEV)
2803 		mnt_flags |= MNT_NODEV;
2804 	if (flags & MS_NOEXEC)
2805 		mnt_flags |= MNT_NOEXEC;
2806 	if (flags & MS_NOATIME)
2807 		mnt_flags |= MNT_NOATIME;
2808 	if (flags & MS_NODIRATIME)
2809 		mnt_flags |= MNT_NODIRATIME;
2810 	if (flags & MS_STRICTATIME)
2811 		mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2812 	if (flags & SB_RDONLY)
2813 		mnt_flags |= MNT_READONLY;
2814 
2815 	/* The default atime for remount is preservation */
2816 	if ((flags & MS_REMOUNT) &&
2817 	    ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
2818 		       MS_STRICTATIME)) == 0)) {
2819 		mnt_flags &= ~MNT_ATIME_MASK;
2820 		mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK;
2821 	}
2822 
2823 	sb_flags = flags & (SB_RDONLY |
2824 			    SB_SYNCHRONOUS |
2825 			    SB_MANDLOCK |
2826 			    SB_DIRSYNC |
2827 			    SB_SILENT |
2828 			    SB_POSIXACL |
2829 			    SB_I_VERSION);
2830 
2831 	if (flags & MS_REMOUNT)
2832 		retval = do_remount(&path, flags, sb_flags, mnt_flags,
2833 				    data_page);
2834 	else if (flags & MS_BIND)
2835 		retval = do_loopback(&path, dev_name, flags & MS_REC);
2836 	else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2837 		retval = do_change_type(&path, flags);
2838 	else if (flags & MS_MOVE)
2839 		retval = do_move_mount(&path, dev_name);
2840 	else
2841 		retval = do_new_mount(&path, type_page, sb_flags, mnt_flags,
2842 				      dev_name, data_page);
2843 dput_out:
2844 	path_put(&path);
2845 	return retval;
2846 }
2847 
2848 static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
2849 {
2850 	return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
2851 }
2852 
2853 static void dec_mnt_namespaces(struct ucounts *ucounts)
2854 {
2855 	dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
2856 }
2857 
2858 static void free_mnt_ns(struct mnt_namespace *ns)
2859 {
2860 	ns_free_inum(&ns->ns);
2861 	dec_mnt_namespaces(ns->ucounts);
2862 	put_user_ns(ns->user_ns);
2863 	kfree(ns);
2864 }
2865 
2866 /*
2867  * Assign a sequence number so we can detect when we attempt to bind
2868  * mount a reference to an older mount namespace into the current
2869  * mount namespace, preventing reference counting loops.  A 64bit
2870  * number incrementing at 10Ghz will take 12,427 years to wrap which
2871  * is effectively never, so we can ignore the possibility.
2872  */
2873 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2874 
2875 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2876 {
2877 	struct mnt_namespace *new_ns;
2878 	struct ucounts *ucounts;
2879 	int ret;
2880 
2881 	ucounts = inc_mnt_namespaces(user_ns);
2882 	if (!ucounts)
2883 		return ERR_PTR(-ENOSPC);
2884 
2885 	new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2886 	if (!new_ns) {
2887 		dec_mnt_namespaces(ucounts);
2888 		return ERR_PTR(-ENOMEM);
2889 	}
2890 	ret = ns_alloc_inum(&new_ns->ns);
2891 	if (ret) {
2892 		kfree(new_ns);
2893 		dec_mnt_namespaces(ucounts);
2894 		return ERR_PTR(ret);
2895 	}
2896 	new_ns->ns.ops = &mntns_operations;
2897 	new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2898 	atomic_set(&new_ns->count, 1);
2899 	new_ns->root = NULL;
2900 	INIT_LIST_HEAD(&new_ns->list);
2901 	init_waitqueue_head(&new_ns->poll);
2902 	new_ns->event = 0;
2903 	new_ns->user_ns = get_user_ns(user_ns);
2904 	new_ns->ucounts = ucounts;
2905 	new_ns->mounts = 0;
2906 	new_ns->pending_mounts = 0;
2907 	return new_ns;
2908 }
2909 
2910 __latent_entropy
2911 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2912 		struct user_namespace *user_ns, struct fs_struct *new_fs)
2913 {
2914 	struct mnt_namespace *new_ns;
2915 	struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2916 	struct mount *p, *q;
2917 	struct mount *old;
2918 	struct mount *new;
2919 	int copy_flags;
2920 
2921 	BUG_ON(!ns);
2922 
2923 	if (likely(!(flags & CLONE_NEWNS))) {
2924 		get_mnt_ns(ns);
2925 		return ns;
2926 	}
2927 
2928 	old = ns->root;
2929 
2930 	new_ns = alloc_mnt_ns(user_ns);
2931 	if (IS_ERR(new_ns))
2932 		return new_ns;
2933 
2934 	namespace_lock();
2935 	/* First pass: copy the tree topology */
2936 	copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2937 	if (user_ns != ns->user_ns)
2938 		copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2939 	new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2940 	if (IS_ERR(new)) {
2941 		namespace_unlock();
2942 		free_mnt_ns(new_ns);
2943 		return ERR_CAST(new);
2944 	}
2945 	new_ns->root = new;
2946 	list_add_tail(&new_ns->list, &new->mnt_list);
2947 
2948 	/*
2949 	 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2950 	 * as belonging to new namespace.  We have already acquired a private
2951 	 * fs_struct, so tsk->fs->lock is not needed.
2952 	 */
2953 	p = old;
2954 	q = new;
2955 	while (p) {
2956 		q->mnt_ns = new_ns;
2957 		new_ns->mounts++;
2958 		if (new_fs) {
2959 			if (&p->mnt == new_fs->root.mnt) {
2960 				new_fs->root.mnt = mntget(&q->mnt);
2961 				rootmnt = &p->mnt;
2962 			}
2963 			if (&p->mnt == new_fs->pwd.mnt) {
2964 				new_fs->pwd.mnt = mntget(&q->mnt);
2965 				pwdmnt = &p->mnt;
2966 			}
2967 		}
2968 		p = next_mnt(p, old);
2969 		q = next_mnt(q, new);
2970 		if (!q)
2971 			break;
2972 		while (p->mnt.mnt_root != q->mnt.mnt_root)
2973 			p = next_mnt(p, old);
2974 	}
2975 	namespace_unlock();
2976 
2977 	if (rootmnt)
2978 		mntput(rootmnt);
2979 	if (pwdmnt)
2980 		mntput(pwdmnt);
2981 
2982 	return new_ns;
2983 }
2984 
2985 /**
2986  * create_mnt_ns - creates a private namespace and adds a root filesystem
2987  * @mnt: pointer to the new root filesystem mountpoint
2988  */
2989 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2990 {
2991 	struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2992 	if (!IS_ERR(new_ns)) {
2993 		struct mount *mnt = real_mount(m);
2994 		mnt->mnt_ns = new_ns;
2995 		new_ns->root = mnt;
2996 		new_ns->mounts++;
2997 		list_add(&mnt->mnt_list, &new_ns->list);
2998 	} else {
2999 		mntput(m);
3000 	}
3001 	return new_ns;
3002 }
3003 
3004 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
3005 {
3006 	struct mnt_namespace *ns;
3007 	struct super_block *s;
3008 	struct path path;
3009 	int err;
3010 
3011 	ns = create_mnt_ns(mnt);
3012 	if (IS_ERR(ns))
3013 		return ERR_CAST(ns);
3014 
3015 	err = vfs_path_lookup(mnt->mnt_root, mnt,
3016 			name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3017 
3018 	put_mnt_ns(ns);
3019 
3020 	if (err)
3021 		return ERR_PTR(err);
3022 
3023 	/* trade a vfsmount reference for active sb one */
3024 	s = path.mnt->mnt_sb;
3025 	atomic_inc(&s->s_active);
3026 	mntput(path.mnt);
3027 	/* lock the sucker */
3028 	down_write(&s->s_umount);
3029 	/* ... and return the root of (sub)tree on it */
3030 	return path.dentry;
3031 }
3032 EXPORT_SYMBOL(mount_subtree);
3033 
3034 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3035 		char __user *, type, unsigned long, flags, void __user *, data)
3036 {
3037 	int ret;
3038 	char *kernel_type;
3039 	char *kernel_dev;
3040 	void *options;
3041 
3042 	kernel_type = copy_mount_string(type);
3043 	ret = PTR_ERR(kernel_type);
3044 	if (IS_ERR(kernel_type))
3045 		goto out_type;
3046 
3047 	kernel_dev = copy_mount_string(dev_name);
3048 	ret = PTR_ERR(kernel_dev);
3049 	if (IS_ERR(kernel_dev))
3050 		goto out_dev;
3051 
3052 	options = copy_mount_options(data);
3053 	ret = PTR_ERR(options);
3054 	if (IS_ERR(options))
3055 		goto out_data;
3056 
3057 	ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3058 
3059 	kfree(options);
3060 out_data:
3061 	kfree(kernel_dev);
3062 out_dev:
3063 	kfree(kernel_type);
3064 out_type:
3065 	return ret;
3066 }
3067 
3068 /*
3069  * Return true if path is reachable from root
3070  *
3071  * namespace_sem or mount_lock is held
3072  */
3073 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3074 			 const struct path *root)
3075 {
3076 	while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3077 		dentry = mnt->mnt_mountpoint;
3078 		mnt = mnt->mnt_parent;
3079 	}
3080 	return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3081 }
3082 
3083 bool path_is_under(const struct path *path1, const struct path *path2)
3084 {
3085 	bool res;
3086 	read_seqlock_excl(&mount_lock);
3087 	res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3088 	read_sequnlock_excl(&mount_lock);
3089 	return res;
3090 }
3091 EXPORT_SYMBOL(path_is_under);
3092 
3093 /*
3094  * pivot_root Semantics:
3095  * Moves the root file system of the current process to the directory put_old,
3096  * makes new_root as the new root file system of the current process, and sets
3097  * root/cwd of all processes which had them on the current root to new_root.
3098  *
3099  * Restrictions:
3100  * The new_root and put_old must be directories, and  must not be on the
3101  * same file  system as the current process root. The put_old  must  be
3102  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3103  * pointed to by put_old must yield the same directory as new_root. No other
3104  * file system may be mounted on put_old. After all, new_root is a mountpoint.
3105  *
3106  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3107  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
3108  * in this situation.
3109  *
3110  * Notes:
3111  *  - we don't move root/cwd if they are not at the root (reason: if something
3112  *    cared enough to change them, it's probably wrong to force them elsewhere)
3113  *  - it's okay to pick a root that isn't the root of a file system, e.g.
3114  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3115  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3116  *    first.
3117  */
3118 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3119 		const char __user *, put_old)
3120 {
3121 	struct path new, old, parent_path, root_parent, root;
3122 	struct mount *new_mnt, *root_mnt, *old_mnt;
3123 	struct mountpoint *old_mp, *root_mp;
3124 	int error;
3125 
3126 	if (!may_mount())
3127 		return -EPERM;
3128 
3129 	error = user_path_dir(new_root, &new);
3130 	if (error)
3131 		goto out0;
3132 
3133 	error = user_path_dir(put_old, &old);
3134 	if (error)
3135 		goto out1;
3136 
3137 	error = security_sb_pivotroot(&old, &new);
3138 	if (error)
3139 		goto out2;
3140 
3141 	get_fs_root(current->fs, &root);
3142 	old_mp = lock_mount(&old);
3143 	error = PTR_ERR(old_mp);
3144 	if (IS_ERR(old_mp))
3145 		goto out3;
3146 
3147 	error = -EINVAL;
3148 	new_mnt = real_mount(new.mnt);
3149 	root_mnt = real_mount(root.mnt);
3150 	old_mnt = real_mount(old.mnt);
3151 	if (IS_MNT_SHARED(old_mnt) ||
3152 		IS_MNT_SHARED(new_mnt->mnt_parent) ||
3153 		IS_MNT_SHARED(root_mnt->mnt_parent))
3154 		goto out4;
3155 	if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3156 		goto out4;
3157 	if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3158 		goto out4;
3159 	error = -ENOENT;
3160 	if (d_unlinked(new.dentry))
3161 		goto out4;
3162 	error = -EBUSY;
3163 	if (new_mnt == root_mnt || old_mnt == root_mnt)
3164 		goto out4; /* loop, on the same file system  */
3165 	error = -EINVAL;
3166 	if (root.mnt->mnt_root != root.dentry)
3167 		goto out4; /* not a mountpoint */
3168 	if (!mnt_has_parent(root_mnt))
3169 		goto out4; /* not attached */
3170 	root_mp = root_mnt->mnt_mp;
3171 	if (new.mnt->mnt_root != new.dentry)
3172 		goto out4; /* not a mountpoint */
3173 	if (!mnt_has_parent(new_mnt))
3174 		goto out4; /* not attached */
3175 	/* make sure we can reach put_old from new_root */
3176 	if (!is_path_reachable(old_mnt, old.dentry, &new))
3177 		goto out4;
3178 	/* make certain new is below the root */
3179 	if (!is_path_reachable(new_mnt, new.dentry, &root))
3180 		goto out4;
3181 	root_mp->m_count++; /* pin it so it won't go away */
3182 	lock_mount_hash();
3183 	detach_mnt(new_mnt, &parent_path);
3184 	detach_mnt(root_mnt, &root_parent);
3185 	if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3186 		new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3187 		root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3188 	}
3189 	/* mount old root on put_old */
3190 	attach_mnt(root_mnt, old_mnt, old_mp);
3191 	/* mount new_root on / */
3192 	attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
3193 	touch_mnt_namespace(current->nsproxy->mnt_ns);
3194 	/* A moved mount should not expire automatically */
3195 	list_del_init(&new_mnt->mnt_expire);
3196 	put_mountpoint(root_mp);
3197 	unlock_mount_hash();
3198 	chroot_fs_refs(&root, &new);
3199 	error = 0;
3200 out4:
3201 	unlock_mount(old_mp);
3202 	if (!error) {
3203 		path_put(&root_parent);
3204 		path_put(&parent_path);
3205 	}
3206 out3:
3207 	path_put(&root);
3208 out2:
3209 	path_put(&old);
3210 out1:
3211 	path_put(&new);
3212 out0:
3213 	return error;
3214 }
3215 
3216 static void __init init_mount_tree(void)
3217 {
3218 	struct vfsmount *mnt;
3219 	struct mnt_namespace *ns;
3220 	struct path root;
3221 	struct file_system_type *type;
3222 
3223 	type = get_fs_type("rootfs");
3224 	if (!type)
3225 		panic("Can't find rootfs type");
3226 	mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
3227 	put_filesystem(type);
3228 	if (IS_ERR(mnt))
3229 		panic("Can't create rootfs");
3230 
3231 	ns = create_mnt_ns(mnt);
3232 	if (IS_ERR(ns))
3233 		panic("Can't allocate initial namespace");
3234 
3235 	init_task.nsproxy->mnt_ns = ns;
3236 	get_mnt_ns(ns);
3237 
3238 	root.mnt = mnt;
3239 	root.dentry = mnt->mnt_root;
3240 	mnt->mnt_flags |= MNT_LOCKED;
3241 
3242 	set_fs_pwd(current->fs, &root);
3243 	set_fs_root(current->fs, &root);
3244 }
3245 
3246 void __init mnt_init(void)
3247 {
3248 	int err;
3249 
3250 	mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
3251 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
3252 
3253 	mount_hashtable = alloc_large_system_hash("Mount-cache",
3254 				sizeof(struct hlist_head),
3255 				mhash_entries, 19,
3256 				HASH_ZERO,
3257 				&m_hash_shift, &m_hash_mask, 0, 0);
3258 	mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
3259 				sizeof(struct hlist_head),
3260 				mphash_entries, 19,
3261 				HASH_ZERO,
3262 				&mp_hash_shift, &mp_hash_mask, 0, 0);
3263 
3264 	if (!mount_hashtable || !mountpoint_hashtable)
3265 		panic("Failed to allocate mount hash table\n");
3266 
3267 	kernfs_init();
3268 
3269 	err = sysfs_init();
3270 	if (err)
3271 		printk(KERN_WARNING "%s: sysfs_init error: %d\n",
3272 			__func__, err);
3273 	fs_kobj = kobject_create_and_add("fs", NULL);
3274 	if (!fs_kobj)
3275 		printk(KERN_WARNING "%s: kobj create error\n", __func__);
3276 	init_rootfs();
3277 	init_mount_tree();
3278 }
3279 
3280 void put_mnt_ns(struct mnt_namespace *ns)
3281 {
3282 	if (!atomic_dec_and_test(&ns->count))
3283 		return;
3284 	drop_collected_mounts(&ns->root->mnt);
3285 	free_mnt_ns(ns);
3286 }
3287 
3288 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
3289 {
3290 	struct vfsmount *mnt;
3291 	mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data);
3292 	if (!IS_ERR(mnt)) {
3293 		/*
3294 		 * it is a longterm mount, don't release mnt until
3295 		 * we unmount before file sys is unregistered
3296 		*/
3297 		real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
3298 	}
3299 	return mnt;
3300 }
3301 EXPORT_SYMBOL_GPL(kern_mount_data);
3302 
3303 void kern_unmount(struct vfsmount *mnt)
3304 {
3305 	/* release long term mount so mount point can be released */
3306 	if (!IS_ERR_OR_NULL(mnt)) {
3307 		real_mount(mnt)->mnt_ns = NULL;
3308 		synchronize_rcu();	/* yecchhh... */
3309 		mntput(mnt);
3310 	}
3311 }
3312 EXPORT_SYMBOL(kern_unmount);
3313 
3314 bool our_mnt(struct vfsmount *mnt)
3315 {
3316 	return check_mnt(real_mount(mnt));
3317 }
3318 
3319 bool current_chrooted(void)
3320 {
3321 	/* Does the current process have a non-standard root */
3322 	struct path ns_root;
3323 	struct path fs_root;
3324 	bool chrooted;
3325 
3326 	/* Find the namespace root */
3327 	ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
3328 	ns_root.dentry = ns_root.mnt->mnt_root;
3329 	path_get(&ns_root);
3330 	while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
3331 		;
3332 
3333 	get_fs_root(current->fs, &fs_root);
3334 
3335 	chrooted = !path_equal(&fs_root, &ns_root);
3336 
3337 	path_put(&fs_root);
3338 	path_put(&ns_root);
3339 
3340 	return chrooted;
3341 }
3342 
3343 static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new,
3344 				int *new_mnt_flags)
3345 {
3346 	int new_flags = *new_mnt_flags;
3347 	struct mount *mnt;
3348 	bool visible = false;
3349 
3350 	down_read(&namespace_sem);
3351 	list_for_each_entry(mnt, &ns->list, mnt_list) {
3352 		struct mount *child;
3353 		int mnt_flags;
3354 
3355 		if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type)
3356 			continue;
3357 
3358 		/* This mount is not fully visible if it's root directory
3359 		 * is not the root directory of the filesystem.
3360 		 */
3361 		if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
3362 			continue;
3363 
3364 		/* A local view of the mount flags */
3365 		mnt_flags = mnt->mnt.mnt_flags;
3366 
3367 		/* Don't miss readonly hidden in the superblock flags */
3368 		if (sb_rdonly(mnt->mnt.mnt_sb))
3369 			mnt_flags |= MNT_LOCK_READONLY;
3370 
3371 		/* Verify the mount flags are equal to or more permissive
3372 		 * than the proposed new mount.
3373 		 */
3374 		if ((mnt_flags & MNT_LOCK_READONLY) &&
3375 		    !(new_flags & MNT_READONLY))
3376 			continue;
3377 		if ((mnt_flags & MNT_LOCK_ATIME) &&
3378 		    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
3379 			continue;
3380 
3381 		/* This mount is not fully visible if there are any
3382 		 * locked child mounts that cover anything except for
3383 		 * empty directories.
3384 		 */
3385 		list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
3386 			struct inode *inode = child->mnt_mountpoint->d_inode;
3387 			/* Only worry about locked mounts */
3388 			if (!(child->mnt.mnt_flags & MNT_LOCKED))
3389 				continue;
3390 			/* Is the directory permanetly empty? */
3391 			if (!is_empty_dir_inode(inode))
3392 				goto next;
3393 		}
3394 		/* Preserve the locked attributes */
3395 		*new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
3396 					       MNT_LOCK_ATIME);
3397 		visible = true;
3398 		goto found;
3399 	next:	;
3400 	}
3401 found:
3402 	up_read(&namespace_sem);
3403 	return visible;
3404 }
3405 
3406 static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags)
3407 {
3408 	const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
3409 	struct mnt_namespace *ns = current->nsproxy->mnt_ns;
3410 	unsigned long s_iflags;
3411 
3412 	if (ns->user_ns == &init_user_ns)
3413 		return false;
3414 
3415 	/* Can this filesystem be too revealing? */
3416 	s_iflags = mnt->mnt_sb->s_iflags;
3417 	if (!(s_iflags & SB_I_USERNS_VISIBLE))
3418 		return false;
3419 
3420 	if ((s_iflags & required_iflags) != required_iflags) {
3421 		WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
3422 			  required_iflags);
3423 		return true;
3424 	}
3425 
3426 	return !mnt_already_visible(ns, mnt, new_mnt_flags);
3427 }
3428 
3429 bool mnt_may_suid(struct vfsmount *mnt)
3430 {
3431 	/*
3432 	 * Foreign mounts (accessed via fchdir or through /proc
3433 	 * symlinks) are always treated as if they are nosuid.  This
3434 	 * prevents namespaces from trusting potentially unsafe
3435 	 * suid/sgid bits, file caps, or security labels that originate
3436 	 * in other namespaces.
3437 	 */
3438 	return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
3439 	       current_in_userns(mnt->mnt_sb->s_user_ns);
3440 }
3441 
3442 static struct ns_common *mntns_get(struct task_struct *task)
3443 {
3444 	struct ns_common *ns = NULL;
3445 	struct nsproxy *nsproxy;
3446 
3447 	task_lock(task);
3448 	nsproxy = task->nsproxy;
3449 	if (nsproxy) {
3450 		ns = &nsproxy->mnt_ns->ns;
3451 		get_mnt_ns(to_mnt_ns(ns));
3452 	}
3453 	task_unlock(task);
3454 
3455 	return ns;
3456 }
3457 
3458 static void mntns_put(struct ns_common *ns)
3459 {
3460 	put_mnt_ns(to_mnt_ns(ns));
3461 }
3462 
3463 static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns)
3464 {
3465 	struct fs_struct *fs = current->fs;
3466 	struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
3467 	struct path root;
3468 	int err;
3469 
3470 	if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
3471 	    !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
3472 	    !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
3473 		return -EPERM;
3474 
3475 	if (fs->users != 1)
3476 		return -EINVAL;
3477 
3478 	get_mnt_ns(mnt_ns);
3479 	old_mnt_ns = nsproxy->mnt_ns;
3480 	nsproxy->mnt_ns = mnt_ns;
3481 
3482 	/* Find the root */
3483 	err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
3484 				"/", LOOKUP_DOWN, &root);
3485 	if (err) {
3486 		/* revert to old namespace */
3487 		nsproxy->mnt_ns = old_mnt_ns;
3488 		put_mnt_ns(mnt_ns);
3489 		return err;
3490 	}
3491 
3492 	put_mnt_ns(old_mnt_ns);
3493 
3494 	/* Update the pwd and root */
3495 	set_fs_pwd(fs, &root);
3496 	set_fs_root(fs, &root);
3497 
3498 	path_put(&root);
3499 	return 0;
3500 }
3501 
3502 static struct user_namespace *mntns_owner(struct ns_common *ns)
3503 {
3504 	return to_mnt_ns(ns)->user_ns;
3505 }
3506 
3507 const struct proc_ns_operations mntns_operations = {
3508 	.name		= "mnt",
3509 	.type		= CLONE_NEWNS,
3510 	.get		= mntns_get,
3511 	.put		= mntns_put,
3512 	.install	= mntns_install,
3513 	.owner		= mntns_owner,
3514 };
3515